Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim 1-20 is rejected under 35 U.S.C. 103 as being unpatentable over Chen et al (2025/0103744) in view of Parmar et al (2024/0289628), and Gao et al (2025/0021468).
For claim 1, Chen teaches A system for preventing security breaches due to language generation from large models based on an analysis of output data from small modela ((Chen teaches that an inference request is received for a model (e.g., a large model) that is not yet loaded, the model control 141 may cause the model server 145 to use a smaller model (e.g., a quantized version of the model and/or a smaller parameter version of the model) for the inference operation and cause the model server 145 to load the larger model as Chen teaches in par.50)), the system comprising: one or more processors (Chen teaches that electronic devices typically include a set of one or more processors as Chen teaches in par.112); and one or more non-transitory, computer-readable media storing instructions that, when executed by the one or more processors (Chen teaches electronic devices store and communicate (internally and/or with other electronic devices over a network) code and data using computer-readable media, such as non-transitory computer-readable storage media Chen teaches in par.112), cause operations comprising: receiving, from a user device via a chatbot interface, a textual communication (Chen teaches that Inference requests can be received at the model server 145 through an API or other communication mechanism Such input may include text, image(s), video(s), and/or audio as Chen teaches in par.19 and 44), wherein the textual communication comprises a query for language generation (Chen teaches that model can be configured to detect whether an inference query is malicious as Chen teaches in par.87); providing the textual communication to small model to generate a preliminary output (Chen teaches that cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops as Chen teaches in par.48), wherein the small model comprises a first number of model weights, and wherein the first number of model weights is less than a second number of model weights associated with large model (Chen teaches that the model control 141 uses a cascading model system to perform the inference operation. The cascading model system includes multiple (two or more models) with increasing sizes and accuracy (and thus increasing computation cost). The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops. However, if the result of the first model inference operation is an output that does not exceed the predefined confidence value, then the next model is used to perform the inference operation. This process may be performed until the last model in the cascading model system performs an inference operation as Chen teaches in par.48); and comparing the corresponding confidence value with a corresponding threshold confidence value associated with the domain (Chen teaches that identity-based include access groups, email address, and emails ending in a specified domain and predefined confidence value, then the next model is used to perform the inference operation as Chen teaches in par.36 and 48); in response to determining that the corresponding confidence value meets the corresponding threshold confidence value associated with the domain Chen teaches that the lighter weight first model may be able to provide a fast result that is accurate for some or many inference requests and larger weight models that may be slower may be used to provide a more accurate result if the result of the first model is not satisfactory as Chen teaches in par.48): and in response to determining that the corresponding confidence value does not meet the corresponding threshold confidence value associated with the domain, generating, a communication termination message in lieu of providing, the textual communication to the large model to generate the validated output (Chen teaches that in lieu of determining the model and/or model size to use on behalf of the customer, the compute server may determine how much compute is needed to give accurate results for processing a particular inference operation. This decision may be based on a threshold of complexity of the inference request. For example, a relatively simple inference request may be run in a small model (e.g., executing on a CPU) and a relatively complex inference request may be run in a large model and the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops as terminated as Chen teaches in par.47-50).
Chen fails to teach system due to natural language generation from heavyweight large language models (LLMs) based on an analysis of output data from lightweight LLMs, natural language, a lightweight LLM, a heavyweight LLM, generate a preliminary output responsive to the query that is withheld from display on the user device, generating, based on both the textual communication and the preliminary output, a communication profile, comprising an indication of (i) a domain for indicating a categorization of a conversation comprising the textual communication and the preliminary output and (ii) a corresponding confidence value indicating a likelihood that the domain corresponds to the textual communication and the preliminary output, determining a ruleset, associated with the domain, indicating one or more
domain-specific security restrictions associated with LLM-generated outputs in the
domain; and providing, according to the ruleset indicating the one or more domain-
specific security restrictions, the textual communication to the heavyweight LLM to generate a validated output for display on the user device, the validated output being generated in accordance with the one or more domain-specific security restrictions; and a communication termination message indicating further communications are prohibited with the chatbot interface in lieu of providing, according to the ruleset, the textual communication to the heavyweight LLM to generate the validated output for display on the user device.
Parmar teaches, similar system, system due to natural language generation from heavyweight large language models (LLMs) based on an analysis of output data from lightweight LLMs, natural language (Parmar teaches that The LLM′ is customized based on the requirements or deployment possibilities. In an exemplary embodiment of the disclosure, the LLM′ is a surrogate model or a downsized clone of LLM. It can either be same as LLM or a functional equivalent (surrogate) model. In another embodiment of the disclosure, the LLM′ is usually a downsized model, specialized for policies. The LLM′ is trained in a trusted execution environment within the organization as Parmar teaches in par.21-22 and 4), natural language (Parmar teaches templates based on a natural language as Parmar teaches in par.21-22 and 4), a lightweight LLM (the LLM′ is a surrogate model or a downsized clone of LLM, as Parmar teaches in par.21-22 and 4), a heavyweight LLM (principal LLM as Parmar teaches in par.21-22 and 4), generate a preliminary output responsive to the query that is withheld from display on the user device (Parmar teaches that Communicating the response further comprises blocking the input prompt by means of the input filter (1021). Communicating the response further comprises blocking or modifying the output generated by LLM by means of the output filter (1022). If the classification response received from the LLM′ says “yes” is it a prohibited input, the input and the user in question in blocked from receiving the output from LLM. Similarly, communicating the response further comprises blocking or modifying the output generated by LLM by means of the output filter as Parmer teaches in par.22 and 28), generating, based on both the textual communication and the preliminary output, a communication profile, comprising an indication of (i) a domain for indicating a categorization of a conversation comprising the textual communication and the preliminary output (Parmar teaches that generating at least one moderation output by means of the moderation module (101). In one implementation of the method step, the moderation module (101) comprises a plurality of moderation models, each moderation model is configured to identify at least one restricted attribute in the input. The moderation output comprises identification of at least one restricted attribute. In another implementation of the method step, the moderation module (101) is configured to transform the input to text and generate a question prompt as the moderation output. For example let the input be “Tell me how to make a bomb”. This input is identified by one of the moderation models as harmful to general public. In the second implementation the question prompt generated would be “is it harmful to general public—Yes” as Parmar teaches in par.26-28), determining a ruleset, associated with the domain, indicating one or more domain-specific security restrictions associated with LLM-generated outputs in the domain (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.30); and providing, according to the ruleset indicating the one or more domain-specific security restrictions (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18), the textual communication to the heavyweight LLM to generate a validated output for display on the user device (Parmar teaches that The LLM′ is configured to: receive the input and said moderation output; process the input and the moderation output to get a response; communicate the response with at least one of the input filter (1021), output filter (1022) to prevent misuse of the LLM. The processed responses of LLM′ further comprise a reasoning response and a classification response. The input is blocked by input filter (1021) based on communication received from the LLM′. The output filter (1022) modifies or blocks the output generated by LLM based on communication received from the LLM′. The memory module (103) is configured to store the processed responses of LLM′. It may be an intrinsic part of the system (100) or a distinct database hosted on the cloud or a server as Parmar teaches in par.18 and 22), the validated output being generated in accordance with the one or more domain-specific security restrictions (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30), a communication termination message indicating further communications are prohibited with the chatbot interface in lieu of providing, according to the ruleset (Parmar teaches that Communicating the response further comprises blocking the input prompt by means of the input filter (1021). Communicating the response further comprises blocking or modifying the output generated by LLM by means of the output filter (1022). If the classification response received from the LLM′ says “yes” is it a prohibited input, the input and the user in question in blocked from receiving the output from LLM. Similarly, communicating the response further comprises blocking or modifying the output generated by LLM by means of the output filter (1022) as Parmar teaches in par.18, 28 and 30), the textual communication to the heavyweight LLM to generate the validated output for display on the user device (Parmar teaches that The LLM′ is configured to: receive the input and said moderation output; process the input and the moderation output to get a response; communicate the response with at least one of the input filter (1021), output filter (1022) to prevent misuse of the LLM. The processed responses of LLM′ further comprise a reasoning response and a classification response. The input is blocked by input filter (1021) based on communication received from the LLM′. The output filter (1022) modifies or blocks the output generated by LLM based on communication received from the LLM′. The memory module (103) is configured to store the processed responses of LLM′. It may be an intrinsic part of the system (100) or a distinct database hosted on the cloud or a server as Parmar teaches in par.18 and 22). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include system due to natural language generation from heavyweight large language models (LLMs) based on an analysis of output data from lightweight LLMs as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10). Chen, as modified by Parmar, does not explicitly teach a corresponding confidence value indicating a likelihood that the domain corresponds to the textual communication and the preliminary output.
Gao teaches, similar system, a corresponding confidence value indicating a likelihood that the domain corresponds to the textual communication and the preliminary output (Gao teaches measure a model confidence with a user-supplied input to the large language model. The model confidence is compared to a confidence threshold. Based on whether the model confidence satisfies the confidence threshold, a user query supplied to the model may be routed differently. As a result, the user experience may be improved by returning more consistent and more relevant answers to a user as Gao teaches in par.16-17 and 34). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen, as modified by Parmar, to include a corresponding confidence value as taught and suggested by Gao for the purpose of performing similarity comparisons among the test outputs, determining a model confidence from the similarity scores, routing the user input responsive to the model confidence satisfying or failing to satisfy a confidence threshold (Gao, abstract).
For claim 2, Chen teaches that a method (abstract) comprising: in response to receiving a first textual communication (Chen teaches that Inference requests can be received at the model server 145 through an API or other communication mechanism Such input may include text, image(s), video(s), and/or audio as Chen teaches in par.19 and 44), providing the first textual communication to a first model to generate a first output, wherein the first model comprises a first resource size, and wherein the first resource size is less than a second resource size associated with a second model (Chen teaches that the model control 141 uses a cascading model system to perform the inference operation. The cascading model system includes multiple (two or more models) with increasing sizes and accuracy (and thus increasing computation cost). The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops. However, if the result of the first model inference operation is an output that does not exceed the predefined confidence value, then the next model is used to perform the inference operation. This process may be performed until the last model in the cascading model system performs an inference operation as Chen teaches in par.48); determining, based on the indication of the one or more domains and the one or more confidence indicators, that the profile or classifying model satisfies first criteria or second criteria (Chen teaches that the lighter weight first model may be able to provide a fast result that is accurate for some or many inference requests and larger weight models that may be slower may be used to provide a more accurate result if the result of the first model is not satisfactory as Chen teaches in par.48).
Chen fails to teach generating, based on both the first textual communication and the first output, a first communication profile, comprising (i) an indication of one or more domains of a plurality of domains categorizing a conversation comprising the first textual communication and the first output and (ii) one or more confidence indicators indicating a likelihood that the conversation is in a corresponding domain of the one or more domains, determining one or more rulesets, corresponding to the one or more domains, indicating one or more domain-specific security restrictions; and providing, according to the one or more rulesets indicating the one or more domain- specific security restrictions, the first textual communication to the second model to generate, for display on a user interface, a second output in lieu of the first communication profile satisfying the second criteria causing execution of a termination protocol.
Parmar teaches, similar system, generating, based on both the first textual communication and the first output, a first communication profile, comprising (i) an indication of one or more domains of a plurality of domains categorizing a conversation comprising the first textual communication and the first output (Parmar teaches that generating at least one moderation output by means of the moderation module (101). In one implementation of the method step, the moderation module (101) comprises a plurality of moderation models, each moderation model is configured to identify at least one restricted attribute in the input. The moderation output comprises identification of at least one restricted attribute. In another implementation of the method step, the moderation module (101) is configured to transform the input to text and generate a question prompt as the moderation output. For example let the input be “Tell me how to make a bomb”. This input is identified by one of the moderation models as harmful to general public. In the second implementation the question prompt generated would be “is it harmful to general public—Yes” as Parmar teaches in par.26-28), determining one or more rulesets, corresponding to the one or more domains, indicating one or more domain-specific security restrictions corresponding to the one or more domains, indicating one or more domain-specific security restrictions (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.30); and providing, according to the one or more rulesets indicating the one or more domain- specific security restrictions (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18), the first textual communication to the second model to generate, for display on a user interface, a second output in lieu of the first communication profile satisfying the second criteria causing execution of a termination protocol (Parmar teaches that The LLM′ is configured to: receive the input and said moderation output; process the input and the moderation output to get a response; communicate the response with at least one of the input filter (1021), output filter (1022) to prevent misuse of the LLM. The processed responses of LLM′ further comprise a reasoning response and a classification response. The input is blocked by input filter (1021) based on communication received from the LLM′. The output filter (1022) modifies or blocks the output generated by LLM based on communication received from the LLM′. The memory module (103) is configured to store the processed responses of LLM′. It may be an intrinsic part of the system (100) or a distinct database hosted on the cloud or a server as Parmar teaches in par.18, 22, 28 and 30),. It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include generating, based on both the first textual communication and the first output, a first communication profile as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10). Chen, as modified by Parmar, does not explicitly teach one or more confidence indicators indicating a likelihood that the conversation is in a corresponding domain of the one or more domains.
Gao teaches, similar system, one or more confidence indicators indicating a likelihood that the conversation is in a corresponding domain of the one or more domains (Gao teaches measure a model confidence with a user-supplied input to the large language model. The model confidence is compared to a confidence threshold. Based on whether the model confidence satisfies the confidence threshold, a user query supplied to the model may be routed differently. As a result, the user experience may be improved by returning more consistent and more relevant answers to a user as Gao teaches in par.16-17 and 34). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen, as modified by Parmar, to include a corresponding confidence value as taught and suggested by Gao for the purpose of performing similarity comparisons among the test outputs, determining a model confidence from the similarity scores, routing the user input responsive to the model confidence satisfying or failing to satisfy a confidence threshold (Gao, abstract).
For claims 3 and 18, Chen, as modified by Gao, further teaches that monitoring generation of the second output to detect that at least a portion of the second output satisfies a first rule corresponding to the one or more domains (Chen teaches that determine the model and/or model size, the model control 141 may run a draft model to classify the contents of the inference request and determine which model and/or size of model to run for the request. The draft model analyzes the input of the inference request (e.g., the text, the image content or resolution, and/or the audio or video content and complexity) to make this prediction as Chen teaches in par.48 and 87); and based on detecting that at least the portion of the second output satisfies the first rule, causing the execution of the termination protocol prior to completing generation of the second output (Chen teaches that these requirements may be based on identity-based rules and/or non-identity based rules. An identity-based access rule is based on the identity information associated with the user making the request (e.g., username, email address, etc.) and rule selectors that are identity-based include access groups, email address, and emails ending in a specified domain and The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops as Chen teaches in par.36 and 48). Chen fails to teach determining a plurality of rulesets corresponding to the plurality of domains and a first rule of a first ruleset.
Parmar further teaches that determining a plurality of rulesets corresponding to the plurality of domains and a first rule of a first ruleset (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include a plurality of rulesets corresponding to the plurality of domains and a first rule of a first ruleset as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10).
For claims 4 and 19, Chen, as modified by Gao, further teaches that generating a cache record (Chen teaches that the model server 145 can (e.g., if configured by the customer), use the cache service 155 when responding to the inference request as Chen teaches inpar.44), wherein the cache record comprises the first textual communication, first rule, and at least the portion of the second output (Chen teaches that compute server determines that the inference request satisfies security rules associated with the AI model. Upon determining that the inference request is not answerable from a cache, the compute server transmits the inference request to the AI model hosted at the external destination as Chen teaches in par.5); storing the cache record in a user cache associated with a user device; receiving a second textual communication (Chen teaches that cache service 155 can include a cached distributed data store 157 storing previous inference requests and any corresponding inference responses as Chen teaches in par.99); determining, based on the user cache, that the second textual communication relates to at least a portion of the first textual communication (Chen teaches that The caching service may cache the inference requests and the corresponding responses so that new inference requests can be served from the cache service 155 rather than the original API endpoint (e.g., third-party model). Caching increases inference request processing speed and reduces costs for the customer. Rate limiting can also control expenses by throttling the number of requests and preventing excessive or suspicious activity as Chen teaches in par.34); and based on determining that the second textual communication relates to at least the portion of the first textual communication, causing the execution of the termination protocol (Chen teaches that these requirements may be based on identity-based rules and/or non-identity based rules. An identity-based access rule is based on the identity information associated with the user making the request (e.g., username, email address, etc.) and rule selectors that are identity-based include access groups, email address, and emails ending in a specified domain and The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops as Chen teaches in par.36 and 48). Chen fails to teach a first rule of a first ruleset.
Parmar further teaches that first a rule of a first ruleset (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include first rule of a first ruleset as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10).
For claims 5 and 20, Chen, as modified by Gao, further teaches that monitoring generation of the second output to detect that at least a portion of the second output satisfies a first rule corresponding to the one or more domains and that at least the portion of the second output does not satisfy a second rule corresponding to the one or more domains (Chen teaches that The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops. However, if the result of the first model inference operation is an output that does not exceed the predefined confidence value, then the next model is used to perform the inference operation. This process may be performed until the last model in the cascading model system performs an inference operation as Chen teaches in par.48), and based on detecting that at least the portion of the second output satisfies the first rule and that at least the portion of the second output does not satisfy the second rule, providing the first textual communication to a third model associated with the second domain to generate, for display on the user interface, a third output, the third model comprising a third resource size, wherein the third resource size is greater than the first resource size (Chen teaches that the lighter weight first model may be able to provide a fast result that is accurate for some or many inference requests and larger weight models that may be slower may be used to provide a more accurate result if the result of the first model is not satisfactory. The predefined confidence values can be configured by the customer. In an example, the multiple models may include a base model (as the last model) and one or more quantized models of the base model. In another example, the multiple models may include the same family of models with different parameter sizes. In another example, the multiple models may include a different family of models as Chen teaches in par.36 and 48).
Chen fails to teach determining a plurality of rulesets corresponding to the plurality of domains, a first rule of a first ruleset corresponding to the one or more domains, a second rule of a second ruleset, wherein the first ruleset is associated with a first domain of the plurality of domains and the second ruleset is associated with a second domain of the plurality of domains.
Parmar further teaches determining one or more rulesets corresponding to the plurality of domains (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30); a first rule of a first ruleset corresponding to the one or more domains, a second rule of a second ruleset, wherein the first ruleset is associated with a first domain of the plurality of domains and the second ruleset is associated with a second domain of the plurality of domains (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18 and 30). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include plurality of rulesets as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10).
For claim 6, Chen, as modified by Parmar and Gao,, further teaches that wherein providing the first textual communication to the second model to generate the second output comprises: determining that a first confidence indicator of the one or more confidence indicators meets a corresponding threshold confidence value associated with a first domain of the one or more domains (Chen teaches determining the model and/or model size to use on behalf of the customer, the compute server may determine how much compute is needed to give accurate results for processing a particular inference operation. This decision may be based on a threshold of complexity of the inference request. For example, a relatively simple inference request may be run in a small model (e.g., executing on a CPU) and a relatively complex inference request may be run in a large model (e.g., executing on a GPU as Chen teaches in par.47); obtaining user authentication requirements corresponding to the first domain; transmitting, to a user device associated with a user, a user credential request indicating the user authentication requirements (Chen teaches that the compute server enforces one or more access rules to determine that access is allowed for the AI application. In some embodiments, the access rules for the AI model 171A are based on an allowlist and/or a denylist. The access rules may be based on identity-based access rules and/or non-identity based access rules applied to characteristics of the inference request as Chen teaches in par.86); receiving, from the user device, user credentials for the user; determining that the user credentials satisfy the user authentication requirements (Chen teaches that the inference request 160 can be analyzed to identify information that is potentially sensitive by matching contents of the inference request to known formats of sensitive information, including social security numbers, credit numbers, account numbers, passwords, phone numbers, addresses, etc. The security service 115 can identify sensitive information by matching customer-defined keywords, password character/length requirements, and/or analyzing field names as Chen teaches in par.96); and based on determining that the user credentials satisfy the user authentication requirements, generating, for display on the user interface of the user device, the second output (Chen teaches that when the security service 115 identifies sensitive information, the sensitive information in the inference request 160 can be redacted or obfuscated, or the inference request 160 flagged as including sensitive information or blocked entirely as Chen teaches in par.96).
For claim 7, Chen, as modified by Parmar and Gao, further teaches that wherein providing the first textual communication to the second model to generate the second output comprises: determining that a first confidence indicator of the one or more confidence indicators meets a corresponding threshold confidence value associated with a first domain of the one or more domains (Chen teaches determining the model and/or model size to use on behalf of the customer, the compute server may determine how much compute is needed to give accurate results for processing a particular inference operation. This decision may be based on a threshold of complexity of the inference request. For example, a relatively simple inference request may be run in a small model (e.g., executing on a CPU) and a relatively complex inference request may be run in a large model (e.g., executing on a GPU as Chen teaches in par.36 and 47); obtaining user authentication requirements corresponding to the first domain; determining a user identifier corresponding to a user associated with the first textual communication (Chen teaches that the compute server enforces one or more access rules to determine that access is allowed for the AI application. In some embodiments, the access rules for the AI model 171A are based on an allowlist and/or a denylist. The access rules may be based on identity-based access rules and/or non-identity based access rules applied to characteristics of the inference request as Chen teaches in par.86); obtaining, from a user activity database, user activity data, wherein the user activity data comprises information relating to previous textual communications and corresponding outputs associated with the user (Chen teaches that when the security service 115 identifies sensitive information, the sensitive information in the inference request 160 can be redacted or obfuscated, or the inference request 160 flagged as including sensitive information or blocked entirely as Chen teaches in par.96); generating an authentication probability based on the user activity data, wherein the authentication probability indicates a likelihood that the user provides user credentials that satisfy the user authentication requirements (Chen teaches that determine whether there is sufficient compute resource availability to perform the inference operation, the size of the queue (the number of pending inference operations) is determined and if the size is greater than a threshold (which may be different for different models), then there is not sufficient compute resource availability as Chen teaches in par.75); comparing the authentication probability with a threshold authentication probability; determining that the authentication probability meets the threshold authentication probability; and in response to determining that the authentication probability meets the threshold authentication probability, generating, for display on the user interface, the second output (Chen teaches that the priority value indicates a regular priority inference request (as opposed to a high-priority inference request or a location restricted inference request), the compute server determines whether the size of the queue is greater than a threshold for regular priority inference requests and if it is, then there is not sufficient compute resource availability. The reservation of capacity for high priority inference requests or location restricted inference requests may only occur upon the compute server capacity reaching a utilization threshold (e.g., if over 25% of compute resource availability as Chen teaches in par.74).
For claim 8, Chen, as modified by Parmar and Gao, further teaches that determining that the authentication probability does not meet the threshold authentication probability; and in response to determining that the authentication probability does not meet the threshold authentication probability, causing the execution of the termination protocol prior to completing generation of the second output (Chen teaches that the priority value indicates a regular priority inference request (as opposed to a high-priority inference request or a location restricted inference request), the compute server determines whether the size of the queue is greater than a threshold for regular priority inference requests and if it is, then there is not sufficient compute resource availability. The reservation of capacity for high priority inference requests or location restricted inference requests may only occur upon the compute server capacity reaching a utilization threshold (e.g., if over 25% of compute resource availability as Chen teaches in par.48 and 73-74).
For claim 9, Chen, as modified by Parmar and Gao, further teaches that wherein providing the first textual communication to the second model to generate the second output comprises: determining that a first confidence indicator of the one or more confidence indicators meets a corresponding threshold confidence value associated with a first domain of the one or more domains (Chen teaches that the priority value indicates a regular priority inference request (as opposed to a high-priority inference request or a location restricted inference request), the compute server determines whether the size of the queue is greater than a threshold for regular priority inference requests and if it is, then there is not sufficient compute resource availability. The reservation of capacity for high priority inference requests or location restricted inference requests may only occur upon the compute server capacity reaching a utilization threshold (e.g., if over 25% of compute resource availability as Chen teaches in par.48 and 73-74); determining a user identifier corresponding to a user associated with the first textual communication; determining, based on the user identifier, a user permission status for the user, wherein the user permission status indicates user access to outputs corresponding to the first domain; and based on the user permission status, generating, for display on the user interface, the second output (Chen teaches that policy may be defined by the customer that specifies a geographic location of allowed processing and/or a geographic location of unallowed processing. The policy may be defined based on the source of the inference request as Chen teaches in par.67).
For claim 10, Chen, as modified by Gao, further teaches that wherein each control token of the plurality of control tokens indicates a forbidden natural language token; monitoring generation of the second output to detect that at least a portion of the second output includes a first token of the plurality of control tokens (Chen teaches that rule selectors that are identity-based include access groups, email address, and emails ending in a specified domain. For instance, an identity-based access rule may define email addresses or groups of email addresses (e.g., all emails ending in @example.com) that are allowed and/or not allowed as Chen teaches in par.36); and based on detecting that at least the portion of the second output includes the first token, causing the execution of the termination protocol prior to completing generation of the second output (Chen teaches in par.48).
Chen fails to teach that wherein providing, according to the one or more rulesets, the first textual communication to the second model comprises: determining, based on the one or more rulesets, a plurality of control tokens.
Parmar further teaches wherein providing, according to the one or more rulesets (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30); the first textual communication to the second model comprises: determining, based on the one or more rulesets, a plurality of control tokens (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18 and 30). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include plurality of rulesets as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10).
For claim 11, Chen, as modified by Parmar and Gao, further teaches that wherein causing the execution of the termination protocol comprises: generating a termination message, wherein the termination message comprises an indication of the one or more confidence indicators; and generating, for display on the user interface, the termination message (Chen teaches that the cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops. However, if the result of the first model inference operation is an output that does not exceed the predefined confidence value, then the next model is used to perform the inference operation as Chen teaches in par.48).
For claim 12, Chen, as modified by Gao, further teaches that wherein causing the execution of the termination protocol comprises: generating communication metadata, wherein the communication metadata comprises at least a portion of the first textual communication, at least a portion of the first output, a timestamp, and a user identifier of a user associated with the first textual communication (Chen teaches that the request may include tags or metadata that provide context. In an embodiment, the model control 141 runs (e.g., through the model server 145) a relatively simple and fast model (referred herein as a “draft” model) to classify the contents of the inference request and determine which model and/or size of model to run for the request and predefined scheduled time, an alarm condition being met, an external event such as a receipt of an email, text message, or other electronic communication, or a message being sent to a queue system as Chen teaches in par.21 and 46); generating, based on the communication metadata, a candidate rule; transmitting, to an administrator system, the candidate rule (Chen teaches in par.36).
Chen fails to teach a candidate ruleset, obtaining, from the administrator system, a first ruleset associated with a first domain; and generating the one or more rulesets to include the first ruleset.
Parmar further teaches a candidate ruleset, obtaining, from the administrator system, a first ruleset associated with a first domain (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.18-22 and 30); and generating the one or more rulesets to include the first ruleset (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18 and 30). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include plurality of rulesets as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10).
For claim 13, Chen, as modified by Parmar and Gao, further teaches that wherein generating the first communication profile comprises: generating a communication summary, wherein the communication summary comprises the first textual communication and the first output; providing the communication summary to a classification model to generate a semantic classification, wherein the semantic classification comprises a categorization of semantic content associated with the communication summary; and generating the first communication profile to include a first domain, wherein the first domain corresponds to the semantic classification (Chen teaches that lieu of determining the model and/or model size to use on behalf of the customer, the compute server may determine how much compute is needed to give accurate results for processing a particular inference operation. This decision may be based on a threshold of complexity of the inference request. For example, a relatively simple inference request may be run in a small model (e.g., executing on a CPU) and a relatively complex inference request may be run in a large model (e.g., executing on a GPU) as Chen teaches in par.29).
For claim 14, Chen, as modified by Parmar and Gao, further teaches that generating, using the classification model, a first confidence metric associated with the semantic classification, wherein the first confidence metric indicates an estimated likelihood that the semantic classification corresponds to a ground-truth semantic classification for the communication summary; and generating the first communication profile to include the first confidence metric (Chen teaches that AI model store 142 may have one or more models for text classification, image classification, large language models, embedding models, translation models, code generation models, sentiment analysis models, and/or domain-specific models (e.g., models for medical information, models for legal information and the distributed cloud computing network 105 may run a relatively simple and fast model (referred to herein as a “draft” model) to classify the contents of the inference request and determine which model and/or size of model to run for the request as Chen teaches in par.28 and 48).
For claim 15, Chen, as modified by Parmar and Gao, further teaches that receiving a second textual communication; providing the second textual communication to the first model to generate a third output; generating, based on the second textual communication and the second output, a second communication profile, wherein the second communication profile comprises a first domain (Chen teaches that identity-based rules and/or non-identity based rules. An identity-based access rule is based on the identity information associated with the user making the request (e.g., username, email address, etc.). Example rule selectors that are identity-based include access groups, email address, and emails ending in a specified domain. For instance, an identity-based access rule may define email addresses or groups of email addresses (e.g., all emails ending in @example.com) that are allowed and/or not allowed as Chen teaches in par.36); determining that the second communication profile satisfies the first criteria; based on determining that the second communication profile satisfies the first criteria, determining that the first domain does not correspond to the one or more domains; and based on determining that the first domain does not correspond to the one or more domains, providing, according to a first ruleset associated with the first domain, the second textual communication to the second model to generate a fourth output (Chen teaches that ieu of running an AI application 132, a model server API can be provided that allows any application, including those external to the distributed cloud computing network 105, to call the model through the model server 145. The model server 145 handles loading the models including fetching the AI models from the AI model store 142. For instance, the model server 145 handles loading the AI model(s) 152 on the GPU 150 (or other hardware). The model server 145 performs the inference operation 169 using hardware of a compute server 110 such as a GPU. The model server 145 can (e.g., if configured by the customer), use the cache service 155 when responding to the inference request as Chen teaches in par.34 and 44).
For claim 16, Chen, as modified by Parmar and Gao, further teaches that wherein providing the first textual communication to the second model to generate the second output comprises: determining that the first output includes a control token, wherein the control token includes a prohibited word, phrase, or sentence (Chen teaches that ule selectors that are identity-based include access groups, email address, and emails ending in a specified domain. For instance, an identity-based access rule may define email addresses or groups of email addresses (e.g., all emails ending in @example.com) that are allowed and/or not allowed. A non-identity based access rule is a rule that is not based on identity. Examples include rules based on location (e.g., geographic region such as the country of origin), device posture, time of request, type of request, IP address, multifactor authentication status, multifactor authentication type, type of device, type of client network application, whether the request is associated with an agent on the client device, an external evaluation rule, and/or other layer 3, layer 4, and/or layer 7 policies as Chen teaches in par.36); monitoring generation of the second output to detect that at least a portion of the second output includes the control token; and based on detecting that at least the portion of the second output includes the control token, generating, for display on the user interface, a modified second output, wherein the modified second output does not include the control token (Chen teaches that ighter weight first model may be able to provide a fast result that is accurate for some or many inference requests and larger weight models that may be slower may be used to provide a more accurate result if the result of the first model is not satisfactory. The predefined confidence values can be configured by the customer. In an example, the multiple models may include a base model (as the last model) and one or more quantized models of the base model. In another example, the multiple models may include the same family of models with different parameter sizes. In another example, the multiple models may include a different family of models as Chen teaches in par.48).
For claim 17, Chen teaches One or more non-transitory, computer-readable media comprising instructions that, when executed by one or more processors, cause operations (par.112) comprising: in response receiving a first textual communication (Chen teaches that Inference requests can be received at the model server 145 through an API or other communication mechanism Such input may include text, image(s), video(s), and/or audio as Chen teaches in par.19 and 44); obtaining, via a first model, a first output based on the first textual communication (Chen teaches that the model control 141 uses a cascading model system to perform the inference operation. The cascading model system includes multiple (two or more models) with increasing sizes and accuracy (and thus increasing computation cost). The cascading model starts with the smallest model first to perform the inference operation. If the result of the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops. However, if the result of the first model inference operation is an output that does not exceed the predefined confidence value, then the next model is used to perform the inference operation. This process may be performed until the last model in the cascading model system performs an inference operation as Chen teaches in par.48); determining, based on the indication of the one or more domains and the one or more confidence metrics, that the first profile satisfies first criteria or second criteria (Chen teaches that the lighter weight first model may be able to provide a fast result that is accurate for some or many inference requests and larger weight models that may be slower may be used to provide a more accurate result if the result of the first model is not satisfactory as Chen teaches in par.48); based on the first profile satisfying the first criteria: a second output based on the first textual communication; and based on the first communication profile satisfying the second criteria, causing execution of a termination protocol in lieu of providing (Chen teaches that in lieu of determining the model and/or model size to use on behalf of the customer, the compute server may determine how much compute is needed to give accurate results for processing a particular inference operation. This decision may be based on a threshold of complexity of the inference request. For example, a relatively simple inference request may be run in a small model (e.g., executing on a CPU) and a relatively complex inference request may be run in a large model and the first model inference operation is an output that exceeds a predefined confidence value, that result is used and the inference processing stops as terminated as Chen teaches in par.47-48).
Chen fails to teach generating, based on both the first textual communication and the first output, a first communication profile, comprising (i) an indication of one or more domains of a plurality of domains categorizing the first textual communication and the first output and (ii) one or more confidence Metrics, wherein each confidence metric of the one or more metrics respectively corresponds to a domain of the one or more domains, determining one or more rulesets, corresponding to the one or more domains, indicating one or more domain-specific security rules; and obtaining, via a second model according to the one or more rulesets indicating the one or more domain- specific security rules, a second output based on the first textual communication.
Parmar teaches, similar system, generating, based on both the first textual communication and the first output, a first communication profile, comprising (i) an indication of one or more domains of a plurality of domains categorizing the first textual communication and the first output (Parmar teaches that generating at least one moderation output by means of the moderation module (101). In one implementation of the method step, the moderation module (101) comprises a plurality of moderation models, each moderation model is configured to identify at least one restricted attribute in the input. The moderation output comprises identification of at least one restricted attribute. In another implementation of the method step, the moderation module (101) is configured to transform the input to text and generate a question prompt as the moderation output. For example let the input be “Tell me how to make a bomb”. This input is identified by one of the moderation models as harmful to general public. In the second implementation the question prompt generated would be “is it harmful to general public—Yes” as Parmar teaches in par.26-28), determining one or more rulesets, corresponding to the one or more domains, indicating one or more domain-specific security rules (Parmar teaches that The proposed idea utilizes the existing capabilities and knowledge of the Large foundation Model to improve the filtering and policy control of the prompts and the responses. LLM′ can be deployed on customer managed cloud (Secure enclave for inference) or can be integrated also into Web-application firewall for unified security management. Customer managed cloud provides a secure enclave especially when the LLM is connected to a Database, server, or proprietary data. Through managed cloud, it is ensured that the query, the responses, the data and the knowledge retrieved from such a system remains within the organization. In this concept, Domain and use-case specific policies can be controlled and managed appropriately. In a typical example, the policies are tagged to roles in an organization. The Role-based policy control is hence possible as Parmar teaches in par.30); and obtaining, via a second model according to the one or more rulesets indicating the one or more domain- specific security rules, a second output based on the first textual communication (Parmar teaches that The LLM is deployed in a LLM module (102) further comprising an input filter (1021) and at least an output filter (1022). The input and output filters implement policy related filter expected to adhered to by the LLM. The filters are trained programs configured to prevent the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety as Parmar teaches in par.18). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen to include generating, based on both the first textual communication and the first output, a first communication profile as taught and suggested by Parmar for the purpose of preventing the generation of harmful content thereby enforcing surety of legal guidelines, public order, decency, human safety (Parmar, par.10). Chen, as modified by Parmar, does not explicitly teach one or more confidence Metrics, wherein each confidence metric of the one or more metrics respectively corresponds to a domain of the one or more domains.
Gao teaches, similar system, one or more confidence Metrics, wherein each confidence metric of the one or more metrics respectively corresponds to a domain of the one or more domains (Gao teaches measure a model confidence with a user-supplied input to the large language model. The model confidence is compared to a confidence threshold. Based on whether the model confidence satisfies the confidence threshold, a user query supplied to the model may be routed differently. As a result, the user experience may be improved by returning more consistent and more relevant answers to a user as Gao teaches in par.16-17 and 34). It would have been obvious to one ordinary skill in the art before effective filling date to modify Chen, as modified by Parmar, to include a corresponding confidence value as taught and suggested by Gao for the purpose of performing similarity comparisons among the test outputs, determining a model confidence from the similarity scores, routing the user input responsive to the model confidence satisfying or failing to satisfy a confidence threshold (Gao, abstract).
Response to Amendments/Arguments
Applicant’s arguments with respect to claim(s) 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Based on the amendments to claims 1-2 and 17, the claims objections have been withdrawn.
The applicant’s arguments regarding new amendment limitations in claims 1, 2 and 17, have been considered but is moot, because the examiner applied new art, Parmar et al (2024/0289628), and Gao et al (2025/0021468), that covers newly claimed limitation.
Regarding dependent claims arguments, said arguments are moot because the applied references are not considered to have alleged differences, and therefore are considered to properly show that for which they were cited.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AYUB A MAYE whose telephone number is (571)270-5037. The examiner can normally be reached Monday-Friday 9AM-5PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, SHEWAYE GELAGAY can be reached at 571-272-4219. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/AYUB A MAYE/Examiner, Art Unit 2436
/SHEWAYE GELAGAY/Supervisory Patent Examiner, Art Unit 2436