CTNF 18/732,432 CTNF 84223 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Applicant’s filing dated 06/03/2024 has been received and made of record. Claims 1-20 are currently pending in Application 18/732,432. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 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. 07-20-aia AIA 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. 07-20-02-aia AIA This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention. 07-21-aia AIA Claim (s) 1-9, 11-18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over DeWeese (US 2025/0363236 A1) in view of Sohail (US 2023/0063052 A1) . Regarding claims 1 and 11, DeWeese discloses a method/system ( DeWeese : Claim 21, “method”, Claim 40, “system”) comprising: at least one memory device ( DeWeese : Claim 40, “ a memory storage comprising a non-transitory, computer-readable medium comprising instructions; and a hardware-based processor that executes the instructions to carry out stages”) ; and a processing device, operatively coupled with the at least one memory device ( DeWeese : Claim 40, “ a memory storage comprising a non-transitory, computer-readable medium comprising instructions; and a hardware-based processor that executes the instructions to carry out stages”) , to: receive, from a user of an online system, at a chat interface, input ( DeWeese : Paragraphs [0035]-[0036], “a connected AI agent can identify a new electronic message from a user… The message can be an email, instant message…”) ; determine a set of facets for the input, wherein the set of facets is based on data for the user of the online system ( DeWeese : Paragraph [0009], “The selected AI agent can generate query vectors with one or more embedding models. The embedding model or embedding parameters used can correspond to those used to create one or more vector databases that will be searched as part of the AI agent execution. To search a vector database, the agent executor (also called “pipeline engine”) can cause the query vectors to be compared to vectors in the vector database”) ; generate an embedding for the input ( DeWeese : Paragraph [0009], “The selected AI agent can generate query vectors with one or more embedding models. The embedding model or embedding parameters used can correspond to those used to create one or more vector databases that will be searched as part of the AI agent execution. To search a vector database, the agent executor (also called “pipeline engine”) can cause the query vectors to be compared to vectors in the vector database”) ; retrieve a plurality of content item embeddings, wherein each of the plurality of content item embeddings is labeled with one or more facets based on a content item associated with that content item embedding ( DeWeese : Paragraph [0009], “The selected AI agent can generate query vectors with one or more embedding models. The embedding model or embedding parameters used can correspond to those used to create one or more vector databases that will be searched as part of the AI agent execution. To search a vector database, the agent executor (also called “pipeline engine”) can cause the query vectors to be compared to vectors in the vector database”) ; filter the plurality of content item embeddings using the determined set of facets and the labeled one or more facets for each of the plurality of content item embeddings ( DeWeese : Paragraph [0009], “The selected AI agent can generate query vectors with one or more embedding models. The embedding model or embedding parameters used can correspond to those used to create one or more vector databases that will be searched as part of the AI agent execution. To search a vector database, the agent executor (also called “pipeline engine”) can cause the query vectors to be compared to vectors in the vector database”, and Paragraph [0013], “The system can also generate prompts for use with the AI model. These prompts can account for the identified content chunks, can include enterprise prompts, and can be selected based on a prompt policy. The prompts, chunks, corresponding metadata, and content query can be sent to the AI model. The AI model can be prompted to do things like prioritize the most relevant chunks based on context, briefly summarize relevant content items, determine whether the content items should be attached to a response, and otherwise format the response for the particular response message type. The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”) ; determine a set of relevant content items using the input embedding and the filtered plurality of content item embeddings ( DeWeese : Paragraph [0013], “The system can also generate prompts for use with the AI model. These prompts can account for the identified content chunks, can include enterprise prompts, and can be selected based on a prompt policy. The prompts, chunks, corresponding metadata, and content query can be sent to the AI model. The AI model can be prompted to do things like prioritize the most relevant chunks based on context, briefly summarize relevant content items, determine whether the content items should be attached to a response, and otherwise format the response for the particular response message type. The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”) ; generate a response prompt using the input embedding and the set of relevant content items ( DeWeese : Paragraph [0013], “The system can also generate prompts for use with the AI model. These prompts can account for the identified content chunks, can include enterprise prompts, and can be selected based on a prompt policy. The prompts, chunks, corresponding metadata, and content query can be sent to the AI model. The AI model can be prompted to do things like prioritize the most relevant chunks based on context, briefly summarize relevant content items, determine whether the content items should be attached to a response, and otherwise format the response for the particular response message type. The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”) ; generate a response by applying a generative machine learning model to the response prompt ( DeWeese : Paragraph [0013], “The system can also generate prompts for use with the AI model. These prompts can account for the identified content chunks, can include enterprise prompts, and can be selected based on a prompt policy. The prompts, chunks, corresponding metadata, and content query can be sent to the AI model. The AI model can be prompted to do things like prioritize the most relevant chunks based on context, briefly summarize relevant content items, determine whether the content items should be attached to a response, and otherwise format the response for the particular response message type. The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”) ; and send the generated response, via the chat interface , to the user of the online system ( DeWeese : Paragraph [0013], “The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”) . DeWeese does not explicitly disclose the chat interface for the user of the online system (although this may be considered implicit or inherent in DeWeese : Paragraphs [0035]-[0036], “a connected AI agent can identify a new electronic message from a user… The message can be an email, instant message…”). Sohail discloses a chat interface for the user of the online system ( Sohail : Figures 6 and 9 and Paragraph [0148], “Example 900 includes conversation participants 902 and 904, who are holding an unstructured conversation via video chat on devices 906 and 908.”). DeWeese and Sohail are analogous art in the same field of endeavor as the instant invention as both are drawn to machine learning content augmentation systems. 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; that is, it would have been obvious to incorporate Sohail ’s user chat interface into the system of DeWeese to allow for convenient real-world implementations. DeWeese-Sohail teaches 2/12. The method of claim 1/system of claim 11, wherein determining the set of relevant content items comprises: performing a similarity search using the input embedding and the filtered plurality of content item embeddings ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”) . DeWeese-Sohail teaches 3/13. The method of claim 2, wherein determining the set of relevant content items further comprises: determining that a content item of the relevant content items includes a chunk identifier ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”) ; identifying one or more additional content items using the chunk identifier ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”) ; and including the one or more additional content items in the set of relevant content items in response to identifying the one or more additional content items ( DeWeese : Figure 1B and Paragraph [0173], “The identified closest vectors are then correlated to the represented content items 674 and chunks 676”) . DeWeese-Sohail teaches 4. The method of claim 1/system of claim 11, further comprising: retrieving a plurality of content items, wherein each of the plurality of content items includes one or more tags ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”; Sohail : Paragraph [0143], “a machine learning model can receive a representation of one or more intents and a vector of the context signal data and determine a corresponding point in a content item embedding space and select content items with similar (above a threshold) mapping in the embedding space. As another example, the machine learning model can receive a representation of one or more intents and a vector of the context signal data, can determine one or more corresponding categories, and can select content items tagged”) ; filtering the plurality of content items using a set of rules ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”; Sohail : Paragraph [0143], “a machine learning model can receive a representation of one or more intents and a vector of the context signal data and determine a corresponding point in a content item embedding space and select content items with similar (above a threshold) mapping in the embedding space. As another example, the machine learning model can receive a representation of one or more intents and a vector of the context signal data, can determine one or more corresponding categories, and can select content items tagged”) ; generating the plurality of content item embeddings using the filtered plurality of content items ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”; Sohail : Paragraph [0143], “a machine learning model can receive a representation of one or more intents and a vector of the context signal data and determine a corresponding point in a content item embedding space and select content items with similar (above a threshold) mapping in the embedding space. As another example, the machine learning model can receive a representation of one or more intents and a vector of the context signal data, can determine one or more corresponding categories, and can select content items tagged”); and labeling each of the plurality of content item embeddings with the one or more facets using the one or more tags for an associated content item ( DeWeese : Paragraph [0032], “The agent executor can then compare the vectorized query and metadata against the vector database of the dataset and identify a number of most similar vectors. The vectors of the vector database can include metadata, such as information describing which content chunks, files, sections, and privileges correspond to the vectors. This metadata can be applied at the time of vectorization by the embedding model, or later based on management policies at the platform”; Sohail : Paragraph [0143], “a machine learning model can receive a representation of one or more intents and a vector of the context signal data and determine a corresponding point in a content item embedding space and select content items with similar (above a threshold) mapping in the embedding space. As another example, the machine learning model can receive a representation of one or more intents and a vector of the context signal data, can determine one or more corresponding categories, and can select content items tagged”). DeWeese-Sohail teaches 5/14. The method of claim 1/system of claim 11, wherein determining the set of facets comprises: retrieving user data for the user of the online system ( DeWeese : Paragraph [0039], “a vector database can be maintained that is specific to a user. Another vector database can be maintained according to an enterprise group to which the user might belong”, and Paragraph [0055], “The agent executor can identify which datasets to search for semantic similarities. This can be just a single vector database associated with the user”) ; and determining the set of facets using the retrieved user data ( DeWeese : Paragraph [0055], “The agent executor can identify which datasets to search for semantic similarities. This can be just a single vector database associated with the user”, and Paragraph [0056], “The agent executor then compares the query vectors to the vector database to identify closest vectors”) . DeWeese-Sohail teaches 6/15. The method of claim 1/system of claim 11, further comprising: classifying an intent for the input, wherein generating the response prompt uses the classified intent ( DeWeese : Paragraph [0013], “The system can also generate prompts for use with the AI model. These prompts can account for the identified content chunks, can include enterprise prompts, and can be selected based on a prompt policy. The prompts, chunks, corresponding metadata, and content query can be sent to the AI model. The AI model can be prompted to do things like prioritize the most relevant chunks based on context, briefly summarize relevant content items, determine whether the content items should be attached to a response, and otherwise format the response for the particular response message type. The result from the AI model can be post-processed by the agent executor and/or connected AI agent. Then the connected AI agent can send a response message to the user and potentially other users associated with the new electronic message”; Sohail : Paragraph [0097], “intents generated at block 506”) . DeWeese-Sohail teaches 7/16. The method of claim 6, wherein classifying the intent for the input comprises: retrieving user data for the user of the online system, wherein classifying the intent uses the user data and the input ( DeWeese : Paragraph [0032], “The embedding model can determine a semantic meaning of the query and output an array of vectors that represent that meaning”; Sohail : Paragraph [0093], “At block 506, process 500 can generate intents for one or more natural language segments obtained through blocks 502 and 504. An intent can be one or more topics, e.g., specified by one or more words, embeddings, or another identifier for semantic meaning”) . DeWeese-Sohail teaches 8/17. The method of claim 1/system of claim 11, wherein filtering the plurality of content item embeddings using the determined set of facets comprises: determining content item embeddings of the plurality of content item embeddings that are associated with the determined set of facets ( DeWeese : Paragraph [0032], “The embedding model can determine a semantic meaning of the query and output an array of vectors that represent that meaning”) ; and retrieving the determined content item embeddings ( DeWeese : Paragraph [0058], “By comparing the query vectors to the vectors of the vector database, a semantic search can be performed based on the query”) . DeWeese-Sohail teaches 9/18. The method of claim 1/system of claim 11, wherein generating the response prompt using the input and the set of relevant content items comprises: generating the response prompt instructing the generative machine learning model to respond to the input, wherein the response prompt includes links to the set of relevant content items for the generative machine learning model to reference ( DeWeese : Paragraph [0107], “The AI model can also provide content items or links to content items ”) . * DeWeese-Sohail teaches *20*11+4. A system comprising: at least one memory device; and a processing device, operatively coupled with the at least one memory device, to: receive, from a user of an online system, at a chat interface, input; determine a set of facets for the input, wherein the set of facets is based on data for the user of the online system; generate an embedding for the input; retrieve a plurality of content items, wherein each of the plurality of content items includes one or more tags; filter the plurality of content items using a set of rules; generate a plurality of content item embeddings using the filtered plurality of content items; label each of the plurality of content item embeddings with one or more facets using the one or more tags for an associated content item; filter the plurality of content item embeddings using the determined set of facets; determine a set of relevant content items using the input embedding and the filtered plurality of content item embeddings; generate a response prompt using the input and the set of relevant content items; generate a response by applying a generative machine learning model to the response prompt; and send the generated response, via the chat interface, to the user of the online system . 07-21-aia AIA Claim (s) 10 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over DeWeese (US 2025/0363236 A1) and Sohail (US 2023/0063052 A1) as applied to claims 1 and 11 above in further view of Guardrails AI (“How to validate LLM responses continuously in real time”) . DeWeese-Sohail teaches 10/19. The method of claim 1/system of claim 11. DeWeese-Sohail does not explicitly disclose dividing the generated response into a plurality of response subdivisions; validating each of the plurality of response subdivisions; and sending each of the plurality of response subdivisions, via the chat interface, to the user of the online system, in response to successfully validating that response subdivision. However, Guardrails AI teaches these features ( Guardrails AI : Page 3, “Typically, Guardrails has to wait for the entire LLM response to arrive before it performs validation. When you enable streaming, Guardrails instead validates each valid fragment as the LLM returns it.”). DeWeese - Sohail and Guardrails AI are analogous art in the same field of endeavor as the instant invention as both are drawn to generative AI systems. 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; that is, it would have been obvious to incorporate Guardrail AI ’s streaming validation feature into the system of DeWeese-Sohail to allow for users to see validated responses in near-real-time as opposed to having to wait . Conclusion 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Boyd (US 2025/0131624 A1) describes a generative LLM system using augmented prompting . Chaturvedi (US 2024/0054298 A1) describes a machine learning system that hierarchically labels intents. Khan (“Intent Classification- Generative AI based Application Architecture 3”) describes an intent classification system for generative AI. Any inquiry concerning this communication or earlier communications from the examiner should be directed to IMAD HUSSAIN whose telephone number is (571)270-3628. The examiner can normally be reached Monday-Friday 0900-1700 ET. 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, Kamal Divecha can be reached at (571) 272-5863. 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. /IMAD HUSSAIN/Primary Examiner, Art Unit 2453 Application/Control Number: 18/732,432 Page 2 Art Unit: 2453 Application/Control Number: 18/732,432 Page 3 Art Unit: 2453 Application/Control Number: 18/732,432 Page 4 Art Unit: 2453 Application/Control Number: 18/732,432 Page 5 Art Unit: 2453 Application/Control Number: 18/732,432 Page 6 Art Unit: 2453 Application/Control Number: 18/732,432 Page 7 Art Unit: 2453 Application/Control Number: 18/732,432 Page 8 Art Unit: 2453 Application/Control Number: 18/732,432 Page 9 Art Unit: 2453 Application/Control Number: 18/732,432 Page 10 Art Unit: 2453 Application/Control Number: 18/732,432 Page 11 Art Unit: 2453 Application/Control Number: 18/732,432 Page 12 Art Unit: 2453 Application/Control Number: 18/732,432 Page 13 Art Unit: 2453 Application/Control Number: 18/732,432 Page 14 Art Unit: 2453 Application/Control Number: 18/732,432 Page 15 Art Unit: 2453
Prosecution Insights