DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
The following title is suggested: Prompt Generation for Language Model to Route a Natural Language Communication from a User Based on Response Types
The disclosure is objected to because of the following informalities:
In ¶[0046], “a plurality of language model” should be “a plurality of language models”.
In ¶[0072], “contexts may be used” should be “contexts which may be used”.
In ¶[0090], “may be performed convert” should be “may be performed to convert”.
In ¶[0177], “that the human agent support” should be “that the human agent supports”.
Appropriate correction is required.
Election/Restrictions
Applicants’ election without traverse of Group I, Claims 1 to 10, in the reply filed on 27 February 2026 is acknowledged.
Claims 11 to 20 are withdrawn from further consideration pursuant to 37 CFR 1.142(b) as being drawn to a nonelected invention, there being no allowable generic or linking claim. Election was made without traverse in the reply filed on 27 February 2026.
Claim Rejections - 35 USC § 103
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.
Claims 1 to 7 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Sotiriou et al. (U.S. Patent Publication 2025/0069086) in view of Koneru et al. (U.S. Patent Publication 2023/0199116).
Concerning independent claim 1, Sotiriou et al. discloses a system for enhancing agent-customer interactions with a large language model, comprising:
“a communications processing circuit configured to receive a first natural language communication from a user and update a session context” – a context used in generating the customer profile summary may include a real-time conversation transcript, the intent (e.g., objective) of the customer in initiating a communication session; a text-based transcript generated in near real-time may be used as context for a purpose of identifying ideal or preferred agent characteristics or agent traits; a text-based transcript is obtained for an initial communication session (“a session context”) (¶[0024] - ¶[0026]); conversation transcript generator 110 (“a communications processing circuit”) processes interactions between users or customers and agents to create a text-based transcript (¶[0060]: Figure 3); with each new chunk of transcript text, a prompt may be constructed using the latest conversation transcript as context and input to the LLM; as the dialogue continues, additional transcript chunks are processed, allowing the LLM to adjust its responses accordingly; updating the transcript and re-prompting the LLM periodically in this manner allows for dynamic, context-aware responses powered by the most recent conversational details (“update a session context”); implicitly, a conversation includes “a first natural language communication from a user”;
“a prompt generator circuit configured to generate a prompt that includes: a representation of the first natural language communication” – generation of prompts for LLMs incorporate customer profile data and, optionally, a text-based transcript from an initial communication session (¶[0026]); a text-based transcript is then provided as context in a prompt submitted to an LLM as input (¶[0028]: a prompt may be constructed using the latest conversation transcript as context and input to the LLM (¶[0062]: Figure 3); a prompt includes the text-based transcript of the entire communication session between the customer and the agent (¶[0070]: Figure 4); here, text of conversation is “the first natural language communication”;
“and a description of a plurality of language model response types, wherein the description includes a first response type that describes transmitting a reply to a user and a second response type that describes requesting assistance from a human agent” – a text-based conversation transcript, generated in near real-time, may be used as context with a prompt, provided as input to an LLM, for the purpose of identifying ideal or preferred agent characteristics or agent traits; given a conversation that may have occurred with an automated agent or chatbot, a user or customer may be routed to a human agent (“a second response type that describes requesting assistance from a human agent”) who is selected by identifying agent traits or characteristics that are best suited for the specific customer identified using a generative language model or LLM; using the output of the model (e.g., the suggested agent characteristics and traits), an intelligent router of the digital engagement service can query various agent profiles to determine which of several available agents is to be selected to handle the customer communication, based on the agent profile including the desired agent characteristics or traits (¶[0025]); LLM prompt specifically instructs the LLM to analyze the provided context data to identify preferred traits or characteristics of an agent best suited to assist the customer further (¶[0026]); automated agents like virtual chatbots 306 can manage common customer inquiries via text-based messaging and web chats; these virtual agents 306 leverage natural language processing, machine learning, and LLMs to understand text-based requests and respond with relevant answers (“a first response type that describes transmitting a reply to a user”) (¶[0058]: Figure 3); here, “a description of a plurality of language model response types” includes “a first response type” of responding to a request with a chatbot and “a second response type” of routing a request to an agent with specified characteristics; that is, an LLM prompt includes “a description of a plurality of language model response types” including at least “a first response type” to continue a session with a chatbot and “a second response type” to route a request to an human agent;
“a language model configured to receive the prompt and generate a response, wherein the response comprises: a reply to the first natural language communication of the user, or a request for assistance from the human agent” – a generative language model is used to create a contextually relevant customer profile summary of a customer's profile data, where the context used in generating the customer profile summary may include a real-time conversation transcript, the intent (e.g., objective) of the customer in initiating a communication session; customer profile summary can be presented to the agent along with an invitation for the agent to accept an inbound communication request from the customer (“wherein the response comprises: . . . a request for assistance from the human agent”), thereby providing the agent with a better understanding of his or her customer and allowing the agent to provide better overall customer service (¶[0024]); automated agents like virtual chatbots 306 can manage common customer inquiries via text-based messaging and web chats; these virtual agents 306 leverage natural language processing, machine learning, and LLMs to understand text-based requests and respond with relevant answers (“wherein the response comprises: a reply to the first natural language communication of the user”) (¶[0058]: Figure 3);
“a response routing circuit configured to: detect that the response is a reply to the user and route the response to the user, or detect that the response is a request for assistance from the human agent and route the request to the human agent” – given a conversation that may have occurred with an automated agent or chatbot, a user or customer may be routed to a human agent who is selected by identifying agent traits or characteristics that are best suited for the specific customer; an intelligent router of the digital engagement service can query various agent profiles to determine which of several available agents is to be selected to handle the customer communication (¶[0025]); an initial communication session could have been managed by an automated agent such as a chatbot (¶[0026]); following the bot interaction, the customer may choose to escalate the communication by making an inbound communication request 404; this request signifies the customer's desire to speak directly with a human agent for further assistance; in response to this request, the system, at operation 406, dynamically generates a prompt that is submitted as input to an LLM service 106 (¶[0065]: Figure 4).
Concerning independent claim 1, Sotiriou et al. discloses generating a prompt to a large language model to route a request in a communication session to a chatbot or route a request to a human agent, but does not expressly disclose “a feedback circuit configured to: collect feedback on the session context from the human agent, and update the session context using the feedback.” Mainly, Sotiriou et al. is directed to generating a prompt to a large language model to route a request to a human agent having particular characteristics corresponding to a context of a request, but discloses that a chatbot may be initially engaged for generating a response as a reply to a user’s request. Here, Sotiriou et al. discloses a large language model may generate a summary for a human agent to prepare the agent for responding to the user request, but does not “collect feedback on the session context from the human agent” and “update the session context using the feedback.”
Concerning independent claim 1, Koneru et al. teaches a system for handling customer conversations in a contact center that uses machine learning in a virtual assistant platform to generate a response to a customer’s input utterance and output a response to the customer. An agent platform 170 manages transferring a communication session handled using one of a plurality of virtual assistants 164(1)-164(n) to one or more of a plurality of agent devices 130(1)-130(n). Routing engine 172 assigns a customer conversation that requires human agent intervention to an available human agent operating the one or more agent devices 130(1)-130(n) based on one or more assignment parameters. Routing engine 172 may retrieve data regarding skill set and level of human agents at one or more of agent devices 130(1)-130(n). The retrieved data may be used in transferring the customer conversation to one of the available human agents. (¶[0038] - ¶[0040]: Figure 1) Specifically, contact center server 150 outputs a first response to an utterance based on utterance parameters to agent device 130(1), and a human agent at agent device 130(1) may view and modify the utterance parameters using context form 314. Based on the modified utterance parameters, virtual assistant platform 160 of contact center server may re-execute a dialog flow of any of the intents of virtual assistant 164(1). The human agent may modify the intent of the utterance in context form 314, and based on the modified intent the virtual assistant platform 160 may execute the dialog flow of the modified intent. The human agent at agent device 130(1) can identify and add any utterance parameters to context form 314 that the virtual assistant platform 160 failed to capture. The human agent after reviewing the conversation may identify that virtual assistant platform 160 failed to capture an entity, and add the identified entity and entity value to context form 160. The human agent may send a first response suggested as-is to customer device 110(1) or may modify the first response suggested and send the modified first response to customer device 110(1). (¶[0063] - ¶[0067]: Figure 3) Here, a human agent provides feedback on a context form by modifying an intent or identifying entities that might be missed by a virtual agent platform so as to “collect feedback on the session context from the human agent”. Modifying an intent or identifying entities on a context form is equivalent to “updating the session context using the feedback.” Compare Specification, ¶[00181], which describes collecting feedback includes a human agent reviewing the session context to fix a mistake or otherwise improve the quality of the automated support. An objective is to improve a handling of customer interactions in contact centers by enabling agent-modified-information based on agent-identified information. (Abstract; ¶[0005]) It would have been obvious to one having ordinary skill in the art to collect feedback from a human agent to update a session context as taught by Koneru et al. in a method for enhancing agent-customer interactions with a large language model of Sotiriou et al. for a purpose of improving a handling of customer interactions in contact centers by enabling agent-modified-information.
Concerning claims 2, 4, and 6, Koneru et al. teaches that a human agent may modify the intent of the utterance in context form 314, and based on the modified intent the virtual assistant platform 160 may execute the dialog flow of the modified intent. The human agent after reviewing the conversation may identify that virtual assistant platform 160 failed to capture an entity, and add the identified entity and entity value to context form 160. The human agent may send a first response suggested as-is to customer device 110(1) or may modify the first response suggested and send the modified first response to customer device 110(1). (¶[0063] - ¶[0067]: Figure 3) Here, modifying an intent of an utterance to a context form, adding entities and entity values to a context form, and modifying a first response are equivalent to “an interjection from the human agent.” That is, a human agent ‘interjects’ some modified or additional information into the conversational context through context form 160. Compare Specification, ¶[0085], which describes an interjection as a command, or suggestion, or whisper by a human agent to cause an automated agent to provide better support.
Concerning claims 3 and 5, Sotiriou et al. discloses that a user or customer may be routed to a human agent who is selected by identifying agent traits or characteristics that are best suited for the specific customer; using the output of the model (e.g., the suggested agent characteristics and traits), an intelligent router of the digital engagement service can query various agent profiles to determine which of several available agents is to be selected to handle the customer communication, based on the agent profile including the desired agent characteristics or traits (¶[0025]); an LLM is instructed to analyze this context data to recommend an optimal routing path for the customer's request, potentially specifying an agent or an agent type best suited to address the customer's current needs; the LLM may generate output recommending routing the customer's communication request directly to a specialized technical help queue, where agents with the requisite expertise are available to provide targeted assistance; similarly, if event data indicates a customer's recent exploration of new product offerings or promotional deals, the LLM might generate output directing the customer to a sales assistant who can capitalize on the customer's purchasing intent (¶[0066] - ¶[0067]: Figure 4). Sotiriou et al., then, discloses routing a response to one of a plurality of human agents including a first human agent or “a second human agent”. Koneru et al. teaches “the feedback circuit is further configured to: collect feedback on the session context from the second human agent, and update the session context using the feedback” in a same way for “a second human agent” as for “the human agent” of the independent claim.
Concerning claim 7, Koneru et al. teaches that a human agent may modify an intent of an utterance, add entities and entity values, and add or modify any information in context form 314. (¶[0063] - ¶[0067]: Figure 4) Figures 3A to 3C, Figures 6A to 6B, and Figures 8A to 8F illustrate that information added and modified is ‘text’ on context form 314 (“wherein the collected feedback comprises natural language text from the human agent”).
Concerning claim 10, Sotiriou et al. discloses that large language models (LLMs) generate and provide real-time customer profile summaries (¶[0002]); a generative language model is used to create a contextually relevant customer profile summary of a customer's profile data, where the context used in generating the customer profile summary may include a real-time conversation transcript and the intent of the customer in initiating a communication session (“to generate a summary of the session context for the human agent”); a customer profile summary can be presented to the agent along with an invitation for the agent to accept an inbound communication request from the customer, thereby providing the agent with a better understanding of his or her customer and allowing the agent to provide better overall customer service (¶[0024]).
Claims 8 to 9 are rejected under 35 U.S.C. 103 as being unpatentable over Sotiriou et al. (U.S. Patent Publication 2025/0069086) in view of Koneru et al. (U.S. Patent Publication 2023/0199116) as applied to claim 1 above, and further in view of Byrne et al. (U.S. Patent Publication 2024/0290327).
Sotiriou et al. discloses a large language model for routing a request from a customer interacting with a chatbot to an appropriate human agent, and briefly notes that services may be communicatively coupled via a public network and one or more application programming interface (“API”) specifications and applications 1406 invoke API calls 1450 through the software stack and receive messages 1452 in response to the API calls 1450 (¶[0030]: Figure 1; ¶[0140]: Figure 14) However, Sotiriou et al. does not expressly disclose “wherein the description of the plurality of language model response types further includes a third response type that describes performing an API call” or “wherein the response comprises a request to perform an API call.” That is, Sotiriou et al. discloses response types directed to having a chatbot respond to a request or an agent with specified traits to respond to a request, but not a response type directed to performing an API call.
Byrne et al. teaches language model prediction of API call invocations so that, using a language model, a prediction string based on the utterance includes an application programming interface (API) call to invoke a program via an API and updating a conversational context. (Abstract) Language prediction model 210 determines whether a respective prediction string 214 includes an API call 222 or other means for requesting/retrieving data external to language prediction system 150. (¶[0031]: Figure 1) When the prediction string 214 does not include API call 222, prediction controller 220 generates utterance response 226 for digital assistant application 50 to provide to user 10. (¶[0035]: Figure 2) Byrne et al., then, teaches that a language model generates a prediction of response types including that a request from a user includes an API call or a response from a digital assistant without an API call based on a conversational context. An objective is to predict both program invocations and user responses with high accuracy to simplify an architecture and eliminate a need for hand-written rules. (¶[0024]) It would have been obvious to one having ordinary skill in the art to provide a language model to predict response types including a request to perform an API call as taught by Byrne et al. in a method for enhancing agent-customer interactions with a large language model of Sotiriou et al. for a purpose of predicting invocations with high accuracy without a necessity for hand-written rules.
Conclusion
The prior art made of record and not relied upon is considered pertinent to Applicants’ disclosure.
Beaver, Chu et al., Subramanian et al., and Ostrand et al. disclose related prior art.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MARTIN LERNER whose telephone number is (571) 272-7608. The examiner can normally be reached Monday-Thursday 8:30 AM-6:00 PM.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Richemond Dorvil can be reached at (571) 272-7602. 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.
/MARTIN LERNER/Primary Examiner
Art Unit 2658 March 13, 2026