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 .
This action is in reply to the communications filed on May 15, 2026. The Applicant’s Amendment and Request for Reconsideration has been received and entered.
Claims 1, 3-9, 11-17, and 19-20 are currently pending and have been examined. Claims 1, 9, and 17 have been amended. Claims 2, 10, and 18 have been cancelled.
Response to Arguments
Applicant’s amendments necessitated the new grounds of rejection.
Applicant’s arguments have been fully considered but they are not persuasive. Particularly, Applicant’s arguments are directed to the instantly amended claims, and are thus moot in view of the new grounds of rejection.
The Examiner welcomes Applicant to contact Examiner for a telephonic interview to further prosecution.
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 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 of this title, 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 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Lange (US PGP 2023/0259714) in view of Khosla (US PGP 2025/0005051).
As per claim 1, Lange teaches a method comprising:
configuring a set of tools on an interface system, wherein at least one or more tools are each configured to access external data, and wherein the interface system hosts an agent configured to access a large language model (LLM), wherein parameters of the LLM are trained based on a score generated by a loss function and further updated using gradient descent; (Lange: [0104]-[0110]; [0118]-[0121]; [0049]; [0052] (LLM); [0054]; [0056]; [0102]; [0111]-[0112](The GAIN system updates one or more model parameter values, for example weights and/or biases, of the language model, based on the computed loss, according to block 430. As described herein, the GAIN system or another system can apply backpropagation with gradient descent and update the model parameter values accordingly.))
receiving, from a user of a client device, a user query via a chatbot application, wherein the user query relates to an order of the user; (Lange: [0035]; [0039]; [0109]-[0115])
for one or more iterations, performing using the agent: (Lange: [0006]; [0016]; [0018])
at a first iteration, (Lange: [0016]; [0018])
providing a prompt for input to the LLM, the prompt specifying at least one or a combination of the user query, contextual information, the set of tools, and a request to output an action, wherein the LLM is configured as a transformer architecture including one or more attention layers and is coupled to receive one or more input tokens . . .; (Lange: [0109]-[0121]; [0071]; [0014]; [0002] (Language models (LMs) include machine learning models, such as deep neural networks, recurrent neural networks, transformers, etc., trained to learn a probability distribution over sequences of words or tokens. . . . trained on large quantities of text data, for example including billions of words or tokens))
receiving a response for the first iteration generated by executing the LLM on the prompt; (Lange: [0118]-[0121]; [0072])
parsing the response from the LLM to extract a selected action and action inputs for the selected action; and (Lange: [0118]-[0121]; [0073]; [0077]; [0058])
triggering execution of the . . . tool corresponding to the selected action with the action inputs to generate one or more observations, wherein the action inputs include the user query, and wherein the one or more observations identify an API call related to assigning or unassigning orders to a picker user based on relevant portions of the indexed database; (Lange: [0117]-[0121]; [0057]-[0058] (When the state handler 115 receives an API call, the state handler 115 saves the current state of the conversation graph, before navigating to a new node. The state handler maintains a history of previous nodes traversed during a session, as well as information received from the user frontend 110 during the session. The LM 120 can receive the current state of the conversation graph and the history as input, in addition to user input from the user frontend 110. The current state and history can also include any information previously collected from the user frontend, for example in response to earlier prompts sent to the user frontend. In this way, the LM 120 can learn to not navigate to nodes for which information has already been collected.))
at a second iteration, (Lange: [0016]; [0018])
prompting the LLM using at least the observation of the first iteration; and (Lange: [0074]-[0076] (The state handler 115 sends input to the LM 120, as shown in line 235B. The input includes the current state of the conversation graph (ACTION=OrderTShirt) and indicates what parameter values are missing or provided (color=? indicates the value for the color parameter in ordering a T-Shirt is missing)); [0058])
executing the identified API call with action inputs extracted from a response from the LLM for the second iteration, the response including an identifier for the order and an identifier for the picker user; and (Lange: [0074]-[0076] (As shown in line 240B, the LM 120 processes the input provided according to line 235B. In line 245B, the LM 120 outputs an API call to the state handler 115, to advance the conversation graph to a state associated with prompting the user for the color of the shirt. For example, the askUserForColor() API call can cause the state handler 115 to advance to the color node 203A as shown in FIG. 2A. The state handler 115 advances to the color node 203A and sends a question prompt to the user frontend 110 as one of the available predetermined actions associated with the color node 203A.); [0058]; Fig. 5A; [0116]; Fig. 7; [0136]-[0139]; [0143])
at a third iteration, (Lange: [0016]; [0018])
prompting the LLM using a result of executing the API call to verify the execution of the API call is completed; and (Lange: [0077]-[0078] (In response, the LM 120 sends an API call to the state handler 115 to set the parameter value for color to blue (setColor(“blue”)), according to block 270B. Because all of the conditions of the color node 202A are met, the API call received at block 270B also causes the state handler 115 to advance to the size node 203A. If the LM 120 was unable to extract a parameter value matching the prompt, the LM 120 would send another API call, for example to cause the state handler 115 to repeat the prompt for a color.); [0055]))
generating a response for the picker user based on the verification that the order has been assigned or unassigned to the picker user; and (Lange: [0077]-[0078] (In response, the LM 120 sends an API call to the state handler 115 to set the parameter value for color to blue (setColor(“blue”)), according to block 270B. Because all of the conditions of the color node 202A are met, the API call received at block 270B also causes the state handler 115 to advance to the size node 203A. If the LM 120 was unable to extract a parameter value matching the prompt, the LM 120 would send another API call, for example to cause the state handler 115 to repeat the prompt for a color.); [0055]))
generating, based on the one or more observations, a response to the user query; and (Lange: [0121]-[0125]; [0077]-[0078]; [0055])
transmitting the response to the client device to cause display of the response at the client device. (Lange: [0121]-[0125]; [0074]; [0055])
Lange does not explicitly disclose the following known techniques which are taught by Khosla:
creating an indexed database indexing one or more articles and documents from a knowledge base, wherein the one or more tools include a retrieval augmented generation (RAG) tool for identifying portions of the indexed database that are relevant to an input; (Khosla: [0021]-[0022] (The search systems 124 in FIG. 1 can connect to the natural language question answering service 102 via the network 116. The search systems 124 may be a part of (e.g., an offered service of the natural language question answering service 102) or associated with the natural language question answering service 102. . . . The search systems 124 may be a plurality of search systems which can provide, but are not limited to, passages, documents, and QA pairs to the natural language question answering service 102. The natural language question answering service 102 may take that information and formulate an answer to a natural language question related to a network-based service and/or computer domain. For example, a search system 124 may be a data store which contains frequently asked questions (FAQ) (and associated answers) concerning a specific network-based service (e.g., network-based storage or database). . . . The aggregator component 104 may be a machine learning model trained on Retrieval Augmented Generation (RAG) techniques.); [0034]-[0038] (Additionally, the memory 210 may include a dense retriever module 219 for determining the relevant parts of each document retrieved from the search systems 124. The dense retriever module 219 may be created using dense embedding (encoder) models that can aim to capture the most salient semantic parts of each document retrieved (e.g., from search systems 124) and convert them into a fixed-dimensional dense representation that can be used to create a matrix of fixed-dimensional vectors. To perform retrieval, the dense retriever module 219 may transform the natural language question into a similar dense embedding and find the K nearest neighbors from an index matrix.); [0013]; [0073])
. . . encoded as embeddings in latent space and generate one or more output embeddings (Khosla: [0035] (The dense retriever module 219 may be created using dense embedding (encoder) models that can aim to capture the most salient semantic parts of each document retrieved (e.g., from search systems 124) and convert them into a fixed-dimensional dense representation that can be used to create a matrix of fixed-dimensional vectors. To perform retrieval, the dense retriever module 219 may transform the natural language question into a similar dense embedding and find the K nearest neighbors from an index matrix.); [0045]; [0027])
. . . triggering execution of the RAG tool . . . (Khosla: [0013]; [0021]-[0022]; [0073])
This known technique is applicable to the method of Lange as they both share characteristics and capabilities, namely, they are directed to LLMs and API calls.
One of ordinary skill in the art at the time of filing would have recognized that applying the known technique of Khosla would have yielded predictable results and resulted in an improved method. It would have been recognized that applying the technique of Khosla to the teachings of Lange would have yielded predictable results because the level of ordinary skill in the art demonstrated by the references applied shows the ability to incorporate such indexed database, RAG tool, and embeddings in latent space features into similar methods. Further, applying the creating an indexed database indexing one or more articles and documents from a knowledge base, wherein the one or more tools include a retrieval augmented generation (RAG) tool for identifying portions of the indexed database that are relevant to an input, encoding as embeddings in latent space, and generating one or more output embeddings to the teachings of Lange would have been recognized by those of ordinary skill in the art as resulting in an improved method that would instantiate various network-based services that can process client requests for data. (Khosla: Para [0001]-[0002]).
As per claim 3, Lange/Khosla teach wherein the one or more tools include a tool for invoking an application programming interface (API) call to an endpoint, and wherein for at least one iteration, the selected action is triggering the tool to invoke the API call with the action inputs as parameters to the API call. (Lange: [0034]-[0036])
As per claim 4, Lange/Khosla teach wherein the one or more tools include a tool for executing code in a REPL environment, and wherein for at least one iteration, the selected action is triggering the tool to execute code specified in the action inputs within the REPL environment. (Lange: [0033]-[0036])
As per claim 5, Lange/Khosla teach wherein the contextual information includes a user identifier associated with a user of the user query and an order identifier associated with an order of the user. (Lange: [0028]; [0031]-[0033])
As per claim 6, Lange/Khosla teach further comprising: receiving a second user query; determining that the second user query should not be responded to by the chatbot application; and routing the second user query to a live agent. (Lange: [0012]; [0041])
As per claim 7, Lange/Khosla teach wherein the prompt includes a description for each tool in the set of tools, and one or more inputs for each tool in the set of tools. (Lange: [0033]-[0036])
As per claim 8, Lange/Khosla teach wherein accessing the external data comprises accessing data provided by an online system. (Lange: [0003], [0033]-[0036])
As per claims 9 and 11-16, these claims are substantially similar to claims 1 and 3-8, respectively, and are therefore rejected in the same manner as these claims, as set forth above.
As per claims 17 and 19-20, these claims are substantially similar to claims 1 and 3-4, respectively, and are therefore rejected in the same manner as these claims, as set forth above.
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 JENNIFER V LEE whose telephone number is (571)272-4778. The examiner can normally be reached Monday - Friday 9AM - 5PM EST.
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, MARIA-TERESA can be reached at (571)272-6764. 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.
/JENNIFER V LEE/Examiner, Art Unit 3688
/VICTORIA E. FRUNZI/Primary Examiner, Art Unit 3689 8/10/2026
Prosecution Insights