Prosecution Insights
Last updated: November 04, 2026
Application No. 18/775,727

PROMPT ENGINEERING AND AUTOMATED QUALITY ASSESSMENT FOR LARGE LANGUAGE MODELS

Final Rejection §103
Filed
Jul 17, 2024
Priority
Feb 27, 2023 — provisional 63/487,037 +1 more
Examiner
BOGGS JR., JAMES
Art Unit
2657
Tech Center
2600 — Communications
Assignee
Unitedhealth Group Incorporation
OA Round
2 (Final)
63%
Grant Probability
Moderate
3-4
OA Rounds
10m
Est. Remaining
99%
With Interview

Examiner Intelligence

Grants 63% of resolved cases
63%
Career Allowance Rate
79 granted / 125 resolved
+1.2% vs TC avg
Strong +36% interview lift
Without
With
+36.3%
Interview Lift
resolved cases with interview
Typical timeline
3y 2m
Avg Prosecution
24 currently pending
Career history
162
Total Applications
across all art units

Statute-Specific Performance

§101
11.8%
-28.2% vs TC avg
§103
50.3%
+10.3% vs TC avg
§102
15.6%
-24.4% vs TC avg
§112
18.9%
-21.1% vs TC avg
Black line = Tech Center average estimate • Based on career data from 125 resolved cases

Office Action

§103
DETAILED ACTION 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 . Response to Amendment The Amendment filed August 5, 2026, has been entered. Claims 1 – 20 are pending in the application. Response to Arguments Applicant's arguments, filed August 5, 2026, with respect to the rejections of claims 1 – 20 have been fully considered but they are not persuasive. On pages 11-12 of Applicant’s response, Applicant argues “None of the cited references disclose, teach, or suggest at least "generating,...using a machine learning classifier model, a contextual classification for the one or more request text fields," "extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification," and "identifying,...from the refined document subset, a prompt document subset based on the one or more request field embeddings" as recited by amended claim 1. The Office Action cites Glass as disclosing "generating, by the one or more processors and using a machine learning classifier model, a contextual classification for the one or more request text fields." Office Action, p. 4. Glass discusses that "the query is paired with a passage when passed to the re-ranker model because, in order to determine if a passage is relevant for a query, the model examines both the query and the passage." Glass ¶ [0018]. That is, Glass combines passages and queries to determine the relevance of each pair. The output of this operation is a relevance score of a pair, not a "contextual classification for the one or more request text fields." Additionally, the "contextual classification" of amended claim 1 is for "the one or more request text fields," and is not based on passages or documents. Thus, Glass does not disclose "generating,...using a machine learning classifier model, a contextual classification for the one or more request text fields" as recited by amended claim 1.”. However, Glass et al. (US Patent Application Publication No. 2024/0111969), hereinafter Glass, recites, in paragraph 0018, lines 1-14, "The retrieved documents are then each paired with at least a portion of the query, creating sequence-pairs for classification. In one or more embodiments, the query is paired with a passage when passed to the re-ranker model because, in order to determine if a passage is relevant for a query, the model examines both the query and the passage. These sequence-pairs are then classified as relevant or not relevant, and the logits for the relevant class are used by at least one generator as the new scores for the corresponding documents. By way of example, for a re-ranker based on BERT, the logit for each query-passage-pair is calculated by multiplying a learned weight vector by the final representation vector for the classification ([CLS]) vector in the transformer.", disclosing generating a contextual classification for a request text field using a machine learning classifier model, where a re-ranker model based on bidirectional encoder representations from transformers (BERT) reads on a machine learning classifier model, a query reads on a request text field, and determining if a passage is relevant for a query reads on generating a contextual classification for a request text field, under the broadest reasonable interpretation of “a contextual classification”. Pairing retrieved documents with at least a portion of the query to create sequence-pairs for classification is a method of determining if a passage is relevant for a query, and claim 1 does not exclude this method for generating a contextual classification for the one or more request text fields. Therefore, Glass discloses the claim 1 limitation “generating, by the one or more processors and using a machine learning classifier model, a contextual classification for the one or more request text fields”. On page 12 of Applicant’s response, Applicant argues “Turning to the second limitation, the Office Action cites Glass as disclosing "identifying, by the one or more processors and from the initial document subset, a refined document subset based on the contextual classification." Office Action, p. 5. Glass discusses "determining and/or taking the top-K documents, from among the top-N documents, to be used by the generation model. In such an embodiments, the initial retrieval component is used to find N documents, which are then re-ranked and the top-K among these (K<N) are used in generating the output." Glass ¶ [0019]. Glass thus narrows its document set by re-ranking on relevance scores - a fundamentally different operation from the claimed extraction. Re-ranking documents based on relevance is not the same as "extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification." In amended claim 1, the historical text documents are extracted based on a correspondence to the contextual classification, not selected as a top-K subset of documents ranked by relevance scores. Thus, Glass does not disclose "extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification" as recited by amended claim 1.”. However, Glass recites, in paragraph 0019, lines 1-10, "Additionally, and as also further detailed herein in connection with FIG. 2, there may be more documents retrieved by initial retrieval than will be ultimately used by the generator. Accordingly, at least one embodiment can include determining and/or taking the top-K documents, from among the top-N documents, to be used by the generation model (also referred to herein as the generator). In such an embodiment, the initial retrieval component is used to find N documents, which are then re-ranked and the top-K among these (K≤N) are used in generating the output.", and recites, in paragraph 0029, lines 1-13, "FIG. 4 is a diagram illustrating a Re2G model 426 implementation, according to an example embodiment of the invention. Similar to the embodiment depicted in FIG. 2, FIG. 4 depicts a query 403, at least a portion of which is processed by query encoder 406. ANN index 405 and BM25 index 445 are then used to determine and/or identify the top-N passages 407 corresponding to one or more query vectors (generated by query encoder 406) having a sufficiently high inner product with one or more corpus-related passage vectors (such as detailed in connection with FIG. 1, for example). Accordingly, in one or more embodiments, the addition of re-ranker 408 enables multiple retrieval methods (e.g., BM25 index 445) to be combined."; disclosing identifying a refined document subset from the initial document subset based on the contextual classification, where “documents retrieved by initial retrieval” read on the initial document subset, “the top-N documents” read on a refined document subset, and identifying the top passages corresponding to a query vector having a sufficiently high inner product with one or more corpus-related passage vectors reads on identifying a refined document subset based on the contextual classification. Claim 1 has been amended to include the limitation “wherein identifying the refined document subset comprises extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification”. Procopio et al. (US Patent No. 8,886,648), hereinafter Procopio, is cited as teaching the limitation “wherein identifying the refined document subset comprises extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification”. Procopio recites, in column 3, lines 43-53, "In operation 120, a document signature may be generated and stored. A document signature may include, for example, topics relating to a text of a document (e.g., a source document), a weight for each topic, and possibly other information. Topics related to text of a document may be generated using natural language processing (NLP) or a similar method or process. A natural language processor may, for example, receive a document text and generate or determine topics related to the document text. Topics may be, for example, categories, abstract ideas, subjects, things, and/or concepts representing the content or subject matter of a document.", recites in column 5, lines 32-45, "In operation 230, document signatures of a source document and each document in a set of documents may be compared. A document signature of a source document may, for example, be compared to each document signature associated with each document in a set of documents. Each comparison may be between two documents (e.g., a source document and a document in a set of documents). In comparing document signatures, a weight of a topic associated with a source document may be compared to a weight of the same topic in each document within a set of documents, and similarity scores may be generated based on the comparison (e.g., one similarity score may be computed for each pair-wise document comparison between the source document and a document within the set of documents).", and recites, in column 6, lines 39-42, "In operation 240, documents similar to a source document may be identified. A subset of documents that are similar to a source document may be selected from a set of documents based on similarity scores.", disclosing extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification, where a document signature that includes topics relating to a text of a document and the topics are categories representing the content or subject matter of a document reads on a contextual classification, and identifying documents similar to a source document by comparing the document signature of a source document to each document signature associated with each document in a set of documents reads on extracting a historical text document that corresponds to the contextual classification. Therefore, the combination of Glass in view of Procopio discloses the claim 1 limitation “identifying, by the one or more processors and from the initial document subset, a refined document subset based on the contextual classification, wherein identifying the refined document subset comprises extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification”. On pages 12-13 of Applicant’s response, Applicant further argues “Turning to the third limitation, the Office Action cites Glass as disclosing "identifying, by the one or more processors and from the refined document subset, a prompt document subset based on the one or more request field embeddings." Office Action, p. 6. Glass discusses that "ANN index and BM25 index are then used to determine and/or identify the top-N passages corresponding to one or more query vectors having a sufficiently high inner product with one or more corpus-related passage vectors." Glass ¶ [0029]. Glass's ANN/BM25 indexing relates to an initial ranking of documents based on relevance to query vectors. In other words, Glass describes a selection from the entire corpus of documents, not from a "refined document subset." Amended claim 1 recites this identification step as secondary to the initial identification, since the prompt document subset is identified from the already refined document subset. Thus, Glass does not disclose "identifying, by the one or more processors and from the refined document subset, a prompt document subset based on the one or more request field embeddings" as recited by amended claim 1.”. However, Glass recites, in paragraph 0019, lines 1-10, "Additionally, and as also further detailed herein in connection with FIG. 2, there may be more documents retrieved by initial retrieval than will be ultimately used by the generator. Accordingly, at least one embodiment can include determining and/or taking the top-K documents, from among the top-N documents, to be used by the generation model (also referred to herein as the generator). In such an embodiment, the initial retrieval component is used to find N documents, which are then re-ranked and the top-K among these (K≤N) are used in generating the output.", and recites, in paragraph 0029, lines 1-13, "FIG. 4 is a diagram illustrating a Re2G model 426 implementation, according to an example embodiment of the invention. Similar to the embodiment depicted in FIG. 2, FIG. 4 depicts a query 403, at least a portion of which is processed by query encoder 406. ANN index 405 and BM25 index 445 are then used to determine and/or identify the top-N passages 407 corresponding to one or more query vectors (generated by query encoder 406) having a sufficiently high inner product with one or more corpus-related passage vectors (such as detailed in connection with FIG. 1, for example). Accordingly, in one or more embodiments, the addition of re-ranker 408 enables multiple retrieval methods (e.g., BM25 index 445) to be combined.", disclosing identifying a prompt document subset from the refined document subset based on the one or more request field embeddings, where the “top-K documents” read on the prompt document subset, the “top-N documents” read on the refined document subset, “taking the top-K documents, from among the top-N documents” reads on identifying a prompt document subset from the refined document subset, and identifying the top passages corresponding to one or more query vectors having a sufficiently high inner product with one or more corpus-related passage vectors reads on identifying the prompt document subset based on the one or more request field embeddings. ANN index and BM25 index being used to determine the top-N passages reads on identifying the refined document subset, and an additional re-ranker enabling multiple retrieval methods, including BM25 index, to be combined, where the top-K documents are taken from the re-ranked top-N documents, reads on identifying a prompt document subset from the refined document subset. Therefore, Glass discloses the claim 1 limitation “identifying, by the one or more processors and from the refined document subset, a prompt document subset based on the one or more request field embeddings”. Drawings The drawings are objected to because: Figures 5 and 7 contain text that is not legible and text and lines that are not sufficiently dense and dark. All drawings must be made by a process which will give them satisfactory reproduction characteristics. Every line, number, and letter must be durable, clean, black (except for color drawings), sufficiently dense and dark, and uniformly thick and well-defined. Numbers, letters, and reference characters should not be placed upon hatched or shaded surfaces (see 37 CFR 1.84(l) and 37 CFR 1.84(p)). Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as “amended.” If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance. 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 – 2, 4, 11 – 12, 14 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Glass et al. (US Patent Application Publication No. 2024/0111969), hereinafter Glass, in view of Procopio et al. (US Patent No. 8,886,648), hereinafter Procopio. Regarding claim 1, Glass discloses a computer-implemented method (Paragraph 0042, lines 1-5, "An embodiment of the invention or elements thereof can be implemented in the form of an apparatus including a memory and at least one processor that is coupled to the memory and configured to perform exemplary method steps.") comprising: identifying, by one or more processors and from a document data store, an initial document subset for a generative text request that comprises a request to generate a generative text document based on one or more request text fields, wherein the initial document subset comprises a plurality of historical text documents (Paragraph 0011, lines 1-10, "As described herein, at least one embodiment includes automated online knowledge distillation for natural language generation with retrieval and model re-training actions. For example, such an embodiment includes accessing at least one knowledge base, receiving at least one natural language question regarding and/or related to the at least one knowledge base, and generating at least one natural language answer by processing at least a portion of the at least one natural language question using one or more retrieval models."; Paragraph 0017, lines 1-6, "In one or more example embodiments, implementing a Re2G model begins with encoding a query into a vector for DPR retrieval. Documents with representation vectors having the highest inner product with the query vector (e.g., the top-N documents) are retrieved, for example, from an approximate nearest neighbors (ANN) index."; Generating a natural language answer by processing a natural language question reads on a generative text request that comprises a request to generate a generative text document based on one or more request text fields, and retrieving documents with representation vectors having the highest inner product with the query vector reads on identifying an initial document subset for a generative text request, where the initial document subset comprises a plurality of historical text documents.); generating, by the one or more processors and using a machine learning classifier model, a contextual classification for the one or more request text fields (Paragraph 0018, lines 1-14, "The retrieved documents are then each paired with at least a portion of the query, creating sequence-pairs for classification. In one or more embodiments, the query is paired with a passage when passed to the re-ranker model because, in order to determine if a passage is relevant for a query, the model examines both the query and the passage. These sequence-pairs are then classified as relevant or not relevant, and the logits for the relevant class are used by at least one generator as the new scores for the corresponding documents. By way of example, for a re-ranker based on BERT, the logit for each query-passage-pair is calculated by multiplying a learned weight vector by the final representation vector for the classification ([CLS]) vector in the transformer."; Pairing the retrieved documents with at least a portion of the query creating sequence-pairs for classification, and classifying the sequence-pairs as relevant or not relevant for a query, reads on generating a contextual classification for the one or more request text fields.); identifying, by the one or more processors and from the initial document subset, a refined document subset based on the contextual classification (Paragraph 0019, lines 1-10, "Additionally, and as also further detailed herein in connection with FIG. 2, there may be more documents retrieved by initial retrieval than will be ultimately used by the generator. Accordingly, at least one embodiment can include determining and/or taking the top-K documents, from among the top-N documents, to be used by the generation model (also referred to herein as the generator). In such an embodiment, the initial retrieval component is used to find N documents, which are then re-ranked and the top-K among these (K≤N) are used in generating the output."; Paragraph 0029, lines 1-13, "FIG. 4 is a diagram illustrating a Re2G model 426 implementation, according to an example embodiment of the invention. Similar to the embodiment depicted in FIG. 2, FIG. 4 depicts a query 403, at least a portion of which is processed by query encoder 406. ANN index 405 and BM25 index 445 are then used to determine and/or identify the top-N passages 407 corresponding to one or more query vectors (generated by query encoder 406) having a sufficiently high inner product with one or more corpus-related passage vectors (such as detailed in connection with FIG. 1, for example). Accordingly, in one or more embodiments, the addition of re-ranker 408 enables multiple retrieval methods (e.g., BM25 index 445) to be combined."; Finding the top-N documents to be used by the generation model, where identifying the top passages corresponding to a query vector having a sufficiently high inner product with one or more corpus-related passage vectors, reads on identifying a refined document subset from the initial document subset based on the contextual classification.); generating, by the one or more processors and using a machine learning embedding model, one or more request field embeddings respectively corresponding to the one or more request text fields (Paragraph 0017, lines 1-6, "In one or more example embodiments, implementing a Re2G model begins with encoding a query into a vector for DPR retrieval. Documents with representation vectors having the highest inner product with the query vector (e.g., the top-N documents) are retrieved, for example, from an approximate nearest neighbors (ANN) index."; Encoding a query into a vector reads on generating one or more request field embeddings respectively corresponding to the one or more request text fields.); identifying, by the one or more processors and from the refined document subset, a prompt document subset based on the one or more request field embeddings (Paragraph 0019, lines 1-10, "Additionally, and as also further detailed herein in connection with FIG. 2, there may be more documents retrieved by initial retrieval than will be ultimately used by the generator. Accordingly, at least one embodiment can include determining and/or taking the top-K documents, from among the top-N documents, to be used by the generation model (also referred to herein as the generator). In such an embodiment, the initial retrieval component is used to find N documents, which are then re-ranked and the top-K among these (K≤N) are used in generating the output."; Paragraph 0029, lines 1-13, "FIG. 4 is a diagram illustrating a Re2G model 426 implementation, according to an example embodiment of the invention. Similar to the embodiment depicted in FIG. 2, FIG. 4 depicts a query 403, at least a portion of which is processed by query encoder 406. ANN index 405 and BM25 index 445 are then used to determine and/or identify the top-N passages 407 corresponding to one or more query vectors (generated by query encoder 406) having a sufficiently high inner product with one or more corpus-related passage vectors (such as detailed in connection with FIG. 1, for example). Accordingly, in one or more embodiments, the addition of re-ranker 408 enables multiple retrieval methods (e.g., BM25 index 445) to be combined."; Taking the top-K documents from among the top-N documents reads on identifying a prompt document subset from the refined document subset, where the top-K documents read on the prompt document subset, the top-N documents read on the refined document subset, and identifying the top passages corresponding to one or more query vectors having a sufficiently high inner product with one or more corpus-related passage vectors reads on identifying the prompt document subset based on the one or more request field embeddings.); generating, by the one or more processors and using a large language model (LLM), one or more generative text fields using a generative model prompt based on the prompt document subset and the one or more request text fields (Paragraph 0035, lines 1-12, "Step 506 includes generating at least one natural language answer, in response to the input query, by processing a subset of the set of passages in connection with one or more automated knowledge distillation techniques based at least in part on the ranking of the at least a portion of the set of passages. In at least one embodiment, generating at least one natural language answer includes processing the subset of the set of passages using at least one sequence-to-sequence generation model. In such an embodiment, the at least one sequence-to-sequence generation model can include at least one bidirectional auto-regressive transformer."; Generating a natural language answer in response to the input query by processing a subset of the set of passages using at least one sequence-to-sequence generation model, where the at least one sequence-to-sequence generation model includes at least one bidirectional auto-regressive transformer, reads on generating, using a large language model, one or more generative text fields using a generative model prompt based on the prompt document subset and the one or more request text fields.); and providing, by the one or more processors, a request response comprising the generative text document based on the one or more generative text fields (Paragraph 0037, lines 1-12, "Step 508 includes performing one or more automated actions based at least in part on one or more of the ranking of the at least a portion of the set of passages and the at least one generated natural language answer. In one or more embodiments, performing one or more automated actions includes automatically training the first set of one or more artificial intelligence techniques based at least in part on one or more of the ranking of the at least a portion of the set of passages. Additionally or alternatively, performing one or more automated actions can include outputting the at least one generated natural language answer to at least one user device."; Outputting the at least one generated natural language answer to at least one user device reads on providing a request response comprising the generative text document based on the one or more generative text fields.). Glass does not specifically disclose: wherein identifying the refined document subset comprises extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification. Procopio teaches: wherein identifying the refined document subset comprises extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification (Column 3, lines 43-53, "In operation 120, a document signature may be generated and stored. A document signature may include, for example, topics relating to a text of a document (e.g., a source document), a weight for each topic, and possibly other information. Topics related to text of a document may be generated using natural language processing (NLP) or a similar method or process. A natural language processor may, for example, receive a document text and generate or determine topics related to the document text. Topics may be, for example, categories, abstract ideas, subjects, things, and/or concepts representing the content or subject matter of a document."; Column 5, lines 32-45, "In operation 230, document signatures of a source document and each document in a set of documents may be compared. A document signature of a source document may, for example, be compared to each document signature associated with each document in a set of documents. Each comparison may be between two documents (e.g., a source document and a document in a set of documents). In comparing document signatures, a weight of a topic associated with a source document may be compared to a weight of the same topic in each document within a set of documents, and similarity scores may be generated based on the comparison (e.g., one similarity score may be computed for each pair-wise document comparison between the source document and a document within the set of documents)."; Column 6, lines 39-42, "In operation 240, documents similar to a source document may be identified. A subset of documents that are similar to a source document may be selected from a set of documents based on similarity scores."; Storing a document signature that includes topics relating to a text of a document, where the topics are categories representing the content or subject matter of a document, and identifying documents similar to a source document by comparing the document signature of a source document to each document signature associated with each document in a set of documents, reads on extracting, from the initial document subset, a historical text document of the plurality of historical text documents that corresponds to the contextual classification.). Procopio is considered to be analogous to the claimed invention because it is in the same field of document classification. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass to incorporate the teachings of Procopio to store a document signature that includes topics relating to a text of a document, where the topics are categories representing the content or subject matter of a document, and identify documents similar to a source document by comparing the document signature of a source document to each document signature associated with each document in a set of documents. Doing so would allow for providing fast, accurate, and reliable results in response to a query for documents given a topic (Procopio; Column 4, line 64 - Column 5, line 4). Regarding claim 2, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 1. Procopio further teaches: wherein the generative text request comprises a category field that identifies a predefined category type corresponding to the one or more request text fields, and wherein the plurality of historical text documents correspond to the predefined category type (Column 3, lines 43-53, "In operation 120, a document signature may be generated and stored. A document signature may include, for example, topics relating to a text of a document (e.g., a source document), a weight for each topic, and possibly other information. Topics related to text of a document may be generated using natural language processing (NLP) or a similar method or process. A natural language processor may, for example, receive a document text and generate or determine topics related to the document text. Topics may be, for example, categories, abstract ideas, subjects, things, and/or concepts representing the content or subject matter of a document."; Column 5, lines 32-45, "In operation 230, document signatures of a source document and each document in a set of documents may be compared. A document signature of a source document may, for example, be compared to each document signature associated with each document in a set of documents. Each comparison may be between two documents (e.g., a source document and a document in a set of documents). In comparing document signatures, a weight of a topic associated with a source document may be compared to a weight of the same topic in each document within a set of documents, and similarity scores may be generated based on the comparison (e.g., one similarity score may be computed for each pair-wise document comparison between the source document and a document within the set of documents)."; Column 6, lines 39-42, "In operation 240, documents similar to a source document may be identified. A subset of documents that are similar to a source document may be selected from a set of documents based on similarity scores."; Storing a document signature that includes topics relating to a text of a document, where the topics are categories representing the content or subject matter of a document, and identifying documents similar to a source document by comparing the document signature of a source document to each document signature associated with each document in a set of documents, reads on the generative text request comprises a category field that identifies a predefined category type corresponding to the one or more request text fields, where the plurality of historical text documents correspond to the predefined category type.). Procopio is considered to be analogous to the claimed invention because it is in the same field of document classification. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio to further incorporate the teachings of Procopio to store a document signature that includes topics relating to a text of a document, where the topics are categories representing the content or subject matter of a document, and identify documents similar to a source document by comparing the document signature of a source document to each document signature associated with each document in a set of documents. Doing so would allow for providing fast, accurate, and reliable results in response to a query for documents given a topic (Procopio; Column 4, line 64 - Column 5, line 4). Regarding claim 4, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 1. Glass further discloses: wherein: (i) the document data store comprises the plurality of historical text documents and a plurality of historical field embeddings respectively corresponding to the plurality of historical text documents, and (ii) the prompt document subset is based on a plurality of embedding similarity scores between the one or more request field embeddings and the plurality of historical field embeddings (Paragraph 0021, lines 1-15, "FIG. 1 is a diagram illustrating a representation model initial retrieval, according to an example embodiment of the invention. By way of illustration, FIG. 1 depicts a corpus 102 (e.g., a text data and/or language data corpus), at least a portion of which is processed by passage encoder 104. As also depicted in FIG. 1, query 103 is processed by query encoder 106. In one or more embodiments, passage encoder 104 and query encoder 106 include BERT models. Additionally, such an embodiment can include using at least one [CLS] vector, generated by passage encoder 104 to produce a vector representation for the text sequence output by query encoder 106. Accordingly, in one or more embodiments, both the query encoder 106 and the passage encoder 104 produce vectors from the [CLS] vector, such that the query and passage each have a vector."; Paragraph 0022, lines 1-8, "As also depicted in FIG. 1, an ANN index 105 is used to determine and/or identify passage vectors (generated by passage encoder 104) with a sufficiently high (e.g., meeting and/or exceeding a predetermined threshold value) inner product with the query vector (generated by query encoder 106). Based at least in part on such a determination and/or identification, corresponding passages 107 are selected and/or output."; A corpus reads on a plurality of historical text documents, processing the corpus with a passage encoder to generate passage vectors reads on embeddings corresponding to the plurality of historical text documents, and identifying passage vectors with a sufficiently high inner product with the query vector reads on the prompt document subset being based on a plurality of embedding similarity scores between the one or more request field embeddings and the plurality of historical field embeddings.). Regarding claim 11, arguments analogous to claim 1 are applicable. In addition, Glass discloses: a system comprising: one or more processors; and one or more memories storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations (Paragraph 0042, lines 1-5, "An embodiment of the invention or elements thereof can be implemented in the form of an apparatus including a memory and at least one processor that is coupled to the memory and configured to perform exemplary method steps.") comprising the steps of claim 1. Regarding claim 12, arguments analogous to claim 2 are applicable. Regarding claim 14, arguments analogous to claim 4 are applicable. Regarding claim 18, arguments analogous to claim 1 are applicable. In addition, Glass discloses: one or more non-transitory computer-readable media storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations (Paragraph 0042, lines 1-5, "An embodiment of the invention or elements thereof can be implemented in the form of an apparatus including a memory and at least one processor that is coupled to the memory and configured to perform exemplary method steps.") comprising the steps of claim 1. Claims 3 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Glass in view of Procopio, and further in view of Dugan (US Patent No. 10,740,563). Regarding claim 3, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 1, but does not specifically disclose: wherein (i) the document data store comprises the plurality of historical text documents and a plurality of contextual classification labels respectively corresponding to the plurality of historical text documents, and (ii) the machine learning classifier model is previously trained using the plurality of contextual classification labels as a plurality of ground truths. Dugan teaches: wherein: (i) the document data store comprises the plurality of historical text documents and a plurality of contextual classification labels respectively corresponding to the plurality of historical text documents, and (ii) the machine learning classifier model is previously trained using the plurality of contextual classification labels as a plurality of ground truths (Column 2, lines 38-45, "The trainer 104 learns to classify text based on the training data 112. In this embodiment, the trainer 104 learns to classify patent claims according to a particular legal rule. The trainer 104 obtains a training set of claims from the training data 112, where each claim is labeled or categorized according to its associated disposition. The trainer 104 then employs machine learning techniques to train one or more models to classify claims according to the legal rule."; Training a machine learning model to classify patent claims according to a particular legal rule with training data including a training set of claims where each claim is labeled or categorized according to its associated disposition reads on the machine learning classifier model being previously trained using a plurality of contextual classification labels as a plurality of ground truths, where the plurality of contextual classification labels correspond to a plurality of historical text documents.). Dugan is considered to be analogous to the claimed invention because it is in the same field of document classification. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio to incorporate the teachings of Dugan to train a machine learning model to classify patent claims according to a particular legal rule with training data including a training set of claims where each claim is labeled or categorized according to its associated disposition. Doing so would allow for using machine learning to classify patent claims according to legal rules (Dugan; Column 1, line 66 - Column 2, line 1). Regarding claim 13, arguments analogous to claim 3 are applicable. Claims 5 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Glass in view of Procopio, and further in view of Baessler et al. (US Patent No. 11,797,615), hereinafter Baessler. Regarding claim 5, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 4. Glass further discloses: wherein: (i) the prompt document subset is based on a related document threshold indicative of a threshold number of prompt examples for the generative model prompt, (ii) a first portion of the prompt document subset comprises one or more first historical text documents that are associated with one or more highest embedding similarity scores from the plurality of embedding similarity scores (Paragraph 0029, lines 1-13, "FIG. 4 is a diagram illustrating a Re2G model 426 implementation, according to an example embodiment of the invention. Similar to the embodiment depicted in FIG. 2, FIG. 4 depicts a query 403, at least a portion of which is processed by query encoder 406. ANN index 405 and BM25 index 445 are then used to determine and/or identify the top-N passages 407 corresponding to one or more query vectors (generated by query encoder 406) having a sufficiently high inner product with one or more corpus-related passage vectors (such as detailed in connection with FIG. 1, for example). Accordingly, in one or more embodiments, the addition of re-ranker 408 enables multiple retrieval methods (e.g., BM25 index 445) to be combined."; Identifying the top-N passages corresponding to one or more query vectors reads on the prompt document subset being based on a related document threshold indicative of a threshold number of prompt examples, and identifying the top-N passages corresponding to one or more query vectors having a sufficiently high inner product with one or more corpus-related passage vectors reads one or more first historical text documents that are associated with one or more highest embedding similarity scores from the plurality of embedding similarity scores.). Glass in view of Procopio does not specifically disclose: (iii) a second portion of the prompt document subset comprises one or more second historical text documents that are randomly sampled from the refined document subset. Baessler teaches: (iii) a second portion of the prompt document subset comprises one or more second historical text documents that are randomly sampled from the refined document subset (Column 2, lines 5-7, "In one aspect, the invention relates to a method for providing one or more random sample documents from a corpus of documents using a search engine."; Column 2, lines 20-28, "A search query is sent to the search engine for execution using the search index. The search query is for documents of the corpus with time-stamps within the time window. In response to the sending of the search query, a search result is received from the search engine. The search result identifies a set of documents of the corpus with time-stamps within the time window. One of the documents comprised by the received set of documents is selected randomly."; Column 5, lines 7-12, "A search engine is a software system that is designed to carry out a search, e.g., a web search, on a corpus of documents in a systematic way for particular information specified in a search query. Search results may be presented in form of a list identifying documents satisfying search criteria defined by the search query used."; A list identifying documents satisfying search criteria defined by the search query used reads on a refined document subset, and selecting one of the documents randomly reads on one or more second historical text documents that are randomly sampled from the refined document subset.). Baessler is considered to be analogous to the claimed invention because it is in the same field of document searching. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio to incorporate the teachings of Baessler to identify a list of documents satisfying search criteria defined by a search query and selecting one of the documents randomly. Doing so would allow for ensuring that a random sample of documents from a corpus of documents accessed via a search engine is unbiased regarding content or size of the documents (Baessler; Column 4, lines 27-40). Regarding claim 15, arguments analogous to claim 5 are applicable. Claims 6 – 8 and 16 – 17 are rejected under 35 U.S.C. 103 as being unpatentable over Glass in view of Procopio, and further in view of Patil et al. (US Patent No. 12,524,612), hereinafter Patil, and Sellam et al. ("BLEURT: Learning Robust Metrics for Text Generation"), hereinafter Sellam. Regarding claim 6, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 1, but does not specifically disclose further comprising: generating a plurality of evaluation metrics for the LLM based on a comparison between the one or more request text fields and the one or more generative text fields; generating, using a rating simulation model, an inferred human rating score for the one or more generative text fields based on the plurality of evaluation metrics; and training the LLM to maximize the inferred human rating score. Patil teaches: generating a plurality of evaluation metrics for the LLM based on a comparison between the one or more request text fields and the one or more generative text fields (Column 1, lines 29-52, "In general, this disclosure describes techniques for determining a quality of a paraphrase using a single metric that accounts for adequacy of the paraphrase, a level of novelty of the paraphrase, and fluency of the paraphrase. High-quality paraphrases preserve the meaning of the original text. The extent to which a paraphrase preserves the meaning of the original text may be referred to herein as paraphrase “adequacy.” High-quality paraphrases are novel, meaning that they are substantially different from the original text. High-quality paraphrases are also fluent, meaning that the paraphrase is understandable to readers. A paraphrase may have low-quality when the paraphrase is very similar to the original text, when the paraphrase conveys a different meaning than the original text, or when the paraphrase is difficult or awkward to understand. A single metric that accounts for adequacy, novelty, and fluency of a paraphrase may provide a better indication of the quality of a paraphrase as compared with metrics that only consider adequacy, only consider novelty, or only consider fluency. Using a single metric that considers adequacy, novelty, fluency, and in some examples one or more other factors may provide an accurate picture of the quality of the paraphrase, because the single metric takes into account more than one characteristic that determines the quality of the paraphrase."; Determining adequacy, novelty, and fluency of a generated paraphrase compared to the original text reads on generating a plurality of evaluation metrics based on a comparison between the one or more request text fields and the one or more generative text fields.); generating, [using a rating simulation model], an inferred human rating score for the one or more generative text fields based on the plurality of evaluation metrics (Column 1, lines 29-52, "In general, this disclosure describes techniques for determining a quality of a paraphrase using a single metric that accounts for adequacy of the paraphrase, a level of novelty of the paraphrase, and fluency of the paraphrase. High-quality paraphrases preserve the meaning of the original text. The extent to which a paraphrase preserves the meaning of the original text may be referred to herein as paraphrase “adequacy.” High-quality paraphrases are novel, meaning that they are substantially different from the original text. High-quality paraphrases are also fluent, meaning that the paraphrase is understandable to readers. A paraphrase may have low-quality when the paraphrase is very similar to the original text, when the paraphrase conveys a different meaning than the original text, or when the paraphrase is difficult or awkward to understand. A single metric that accounts for adequacy, novelty, and fluency of a paraphrase may provide a better indication of the quality of a paraphrase as compared with metrics that only consider adequacy, only consider novelty, or only consider fluency. Using a single metric that considers adequacy, novelty, fluency, and in some examples one or more other factors may provide an accurate picture of the quality of the paraphrase, because the single metric takes into account more than one characteristic that determines the quality of the paraphrase."; A single metric that accounts for adequacy, novelty, and fluency of a paraphrase reads on an inferred human rating score for the one or more generative text fields based on the plurality of evaluation metrics, where a single metric reads on a score and adequacy, novelty, and fluency read on a plurality of evaluation metrics.); and training the LLM to maximize the inferred human rating score (Column 2, lines 39-42, "In some cases, metric values calculated by the paraphrase quality evaluation system may be used to generate a training set of high-quality paraphrases to train one or more paraphrase models."; Column 9, lines 49-60, "Paraphrase quality evaluation system 44 may be configured to use several different metrics that evaluate quality paraphrases, including but not limited to ROUGEP. Paraphrase generation system 42 may apply one or more paraphrase generation models (e.g., generative pre-trained transformer 2 (GPT-2)) to generate paraphrases. Paraphrase system 10 may use one or more techniques described herein to develop one or more paraphrase models. Paraphrase system 10 may analyze paraphrase evaluation using ROUGEP. Paraphrase quality metrics such as ROUGEP may enhance an analysis of paraphrase model fine-tuning and generation."; A generative pre-trained transformer model reads on a large language model, and using metric values calculated by the paraphrase quality evaluation system to generate a training set of high-quality paraphrases to train one or more paraphrase models reads on training the large language model to maximize the inferred human rating score.). Patil is considered to be analogous to the claimed invention because it is in the same field of natural language generation. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio to incorporate the teachings of Patil to use metric values calculated by the paraphrase quality evaluation system to generate a training set of high-quality paraphrases to train one or more paraphrase models. Doing so would allow for improving a paraphrase generation model by re-training the paraphrase generation model to consistently generate paraphrases having characteristics common to high-quality paraphrases previously generated by the paraphrase generation model (Patil; Column 27, lines 21-30). Glass in view of Procopio and Patil does not specifically disclose: generating, using a rating simulation model, an inferred human rating score for the one or more generative text fields based on the plurality of evaluation metrics. Sellam teaches: generating, using a rating simulation model, an inferred human rating score for the one or more generative text fields based on the plurality of evaluation metrics (Abstract, lines 5-9, "We propose BLEURT, a learned evaluation metric based on BERT that can model human judgments with a few thousand possibly biased training examples."; Section 4.2, lines 11-16, "Automatic Metrics: We create three signals τBLEU, τROUGE, and τBERTscore with sentence BLEU (Papineni et al., 2002), ROUGE (Lin, 2004), and BERTscore (Zhang et al., 2020) respectively (we use precision, recall and F-score for the latter two)."; Training a Bidirectional Encoder Representations from Transformers model to model human judgments reads on generating an inferred human rating score using a rating simulation model, and BLEU and ROUGE metrics read on a plurality of evaluation metrics.). Sellam is considered to be analogous to the claimed invention because it is in the same field of natural language generation. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio and Patil to incorporate the teachings of Sellam to train a Bidirectional Encoder Representations from Transformers model with BLEU and ROUGE metrics to model human judgments. Doing so would allow for modeling human assessment with superior accuracy (Sellam; Section 7, lines 1-6). Regarding claim 7, Glass in view of Procopio, Patil, and Sellam discloses the computer-implemented method as claimed in claim 6. Patil further teaches: wherein the plurality of evaluation metrics comprises a Bilingual Evaluation Understudy (BLEU) metric, a Recall-Oriented Understudy for Gisting Evaluation (ROUGE) metric, and a Metric for Evaluation of Translation with Explicit Ordering (METEOR) metric (Column 9, lines 61-65, "Paraphrase system 10 may use several paraphrase quality metrics to evaluate model generated paraphrases with reference during testing. Metrics such as BLEU, METEOR, ROUGE, and TER may measure an adequacy of one or more paraphrases."). Patil is considered to be analogous to the claimed invention because it is in the same field of natural language generation. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio, Patil, and Sellam to further incorporate the teachings of Patil to use Bilingual Evaluation Understudy (BLEU), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), and Metric for Evaluation of Translation with Explicit Ordering (METEOR) metrics. Doing so would allow for improving a paraphrase generation model by re-training the paraphrase generation model to consistently generate paraphrases having characteristics common to high-quality paraphrases previously generated by the paraphrase generation model (Patil; Column 27, lines 21-30). Regarding claim 8, Glass in view of Procopio, Patil, and Sellam discloses the computer-implemented method as claimed in claim 6. Sellam further teaches: wherein: (i) the rating simulation model is previously trained using a plurality of historical request-generative text field pairs, and (ii) a historical request-generative text field pair of the plurality of historical request-generative text field pairs is associated with a plurality of historical evaluation metrics and a manual label (Abstract, lines 5-9, "We propose BLEURT, a learned evaluation metric based on BERT that can model human judgments with a few thousand possibly biased training examples."; Section 2, lines 1-8, “Define x = ( x i , …, x r ) to be the reference sentence of length r where each x i is a token and let x̃ = ( x ̃ i , ..., x ̃ p ) be a prediction sentence of length p. Let { ( x i ,   x ̃ i ,   y i ) } n = 1 N be a training dataset of size N where y i ∈ R is the human rating that indicates how good x ̃ i is with respect to x i . Given the training data, our goal is to learn a function f : (x, x̃) → y that predicts the human rating.”; Section 4.2, lines 11-16, "Automatic Metrics: We create three signals τBLEU, τROUGE, and τBERTscore with sentence BLEU (Papineni et al., 2002), ROUGE (Lin, 2004), and BERTscore (Zhang et al., 2020) respectively (we use precision, recall and F-score for the latter two)."; A reference sentence reads on a historical request, a prediction sentence reads on a generative text field, and a human rating that indicates how good a prediction sentence is with respect to reference sentence reads on a manual label.). Sellam is considered to be analogous to the claimed invention because it is in the same field of natural language generation. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio, Patil, and Sellam to further incorporate the teachings of Sellam to train a Bidirectional Encoder Representations from Transformers model with BLEU and ROUGE metrics to model human judgments using a training dataset of reference sentences, prediction sentences, and human ratings. Doing so would allow for modeling human assessment with superior accuracy (Sellam; Section 7, lines 1-6). Regarding claim 16, arguments analogous to claim 6 are applicable. Regarding claim 17, arguments analogous to claim 7 are applicable. Claims 9 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Glass in view of Procopio, and further in view of Dodelin et al. (US Patent No. 12,238,049), hereinafter Dodelin. Regarding claim 9, Glass in view of Procopio discloses the computer-implemented method as claimed in claim 1, but does not specifically disclose further comprising: receiving, via an application programming interface (API) call, the generative text request, wherein the API call is initiated from a generative service plug-in associated with a user device; providing, in response to the API call, a request identifier for the generative text request; and storing the generative text request with the request identifier in a processing queue. Dodelin teaches: receiving, via an application programming interface (API) call, the generative text request, wherein the API call is initiated from a generative service plug-in associated with a user device; providing, in response to the API call, a request identifier for the generative text request; and storing the generative text request with the request identifier in a processing queue (Column 8, lines 37-44, "In one example, dialogue management device 118 may receive a request as a result of a user clicking a pre-defined hyperlink embedded in a received email. In this example, after dialogue management device 118 receives the API call from web server 108, dialogue management device 118 may parse the API call to extract the customer identifier and the request identifier from the API call, and generate a command to place in a command queue, as described in detail below."). Dodelin is considered to be analogous to the claimed invention because it is in the same field of application programming interfaces. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio to incorporate the teachings of Dodelin to receive a request via an application programming interface (API) call from a user device, provide a request identifier for the request in response to the API call, and storing the request in a queue. Doing so would allow for establishing a text-message based communication channel between a client device and an organization (Dodelin; Column 5, lines 19-23). Regarding claim 19, arguments analogous to claim 9 are applicable. Claims 10 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Glass in view of Procopio and Dodelin, and further in view of McClory et al. (US Patent No. 10,521,284), hereinafter McClory. Regarding claim 10, Glass in view of Procopio and Dodelin discloses the computer-implemented method as claimed in claim 9, but does not specifically disclose further comprising: storing the generative text document with the request identifier in a completed queue; receiving a status request comprising the request identifier; and in response to the status request, providing the generative text document to the user device. McClory teaches: storing the generative text document with the request identifier in a completed queue; receiving a status request comprising the request identifier; and in response to the status request, providing the generative text document to the user device (Column 1, lines 40-58, "According to an embodiment, a request for a service may be received by a command application programming interface (API) from a client device. A request identifier and a task message specifying a task to be performed by the service may be generated based on the received request. A read cache may be updated with the request identifier, and the task message may be transmitted to a command task queue. A worker process may detect the task message upon transmission of the task message to the command queue and execute the task specified by the task message. Upon completion of the task, a completion message may be transmitted to an event publish queue. In an embodiment, a status of the task specified by the task message may be periodically transmitted to the read queue. A status request including the request identifier may be received by a read API, and the status from the read queue may be retrieved in response to receiving the status request. The retrieved status may then be transmitted to the client device."; Column 34, lines 2-3, "Finally, at stage 616, the formatted requested data may be transmitted to the client device."). McClory is considered to be analogous to the claimed invention because it is in the same field of application programming interfaces. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Glass in view of Procopio and Dodelin to incorporate the teachings of McClory to transmit a message to an event publish queue upon completion of the task, provide a status request including a request identifier, and transmit requested data to the client device. Doing so would allow for processing requests for a deployed service or application (McClory; Column 1, lines 36-29). Regarding claim 20, arguments analogous to claim 10 are applicable. 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 James Boggs whose telephone number is (571)272-2968. The examiner can normally be reached M-F 8:00 AM - 5:00 PM. 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, Daniel Washburn can be reached at (571)272-5551. 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. /JAMES BOGGS/Examiner, Art Unit 2657
Read full office action

Prosecution Timeline

Jul 17, 2024
Application Filed
May 05, 2026
Non-Final Rejection mailed — §103
Aug 05, 2026
Response Filed
Aug 31, 2026
Final Rejection mailed — §103
Oct 07, 2026
Examiner Interview Summary
Oct 07, 2026
Applicant Interview (Telephonic)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12783797
NON-TRANSITORY STORAGE MEDIUM STORING DOCUMENT EXTRACTION PROGRAM IN COMPUTER LANGUAGE PROCESSING, SEMANTICALLY SIMILAR DOCUMENT EXTRACTION METHOD, AND LANGUAGE PROCESSING APPARATUS
2y 4m to grant Granted Nov 03, 2026
Patent 12776300
VOICE GRAFTING USING MACHINE LEARNING
4y 0m to grant Granted Oct 27, 2026
Patent 12775794
GENERATING A PERSONAL CORPUS
4y 0m to grant Granted Oct 27, 2026
Patent 12776332
SYSTEM AND METHOD FOR MASK-BASED NEURAL BEAMFORMING FOR MULTI-CHANNEL SPEECH ENHANCEMENT
3y 2m to grant Granted Oct 27, 2026
Patent 12776322
BIOMETRIC AUTHENTICATION DEVICE, BIOMETRIC AUTHENTICATION METHOD, AND RECORDING MEDIUM
3y 0m to grant Granted Oct 27, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

3-4
Expected OA Rounds
63%
Grant Probability
99%
With Interview (+36.3%)
3y 2m (~10m remaining)
Median Time to Grant
Moderate
PTA Risk
Based on 125 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month