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 .
This action is responsive to application filed on 2/5/25. Claims 1-20 are presented for examination.
Abstract analysis: The method of embedding queries and targeted documents to query large language model comprises practical application in AI computations and searching and is therefore compliant.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 1-20 are is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Krishnan et al (USPN. 2023/0306087).
Regarding claims 1, 15 and 20, Krishnan discloses a method, apparatus and program comprising non- transitory computer readable storage medium storing instructions, for efficient handling of queries, the method comprising (fig. 1):
receiving, by communications hardware, a query from a user device (fig. 1, par. 29, query input, may be received by the server);
generating, by analysis circuitry, an embedding representation of the query (fig. 1B and 1C, par. 51, query is encoded into different embeddings);
performing, by the analysis circuitry, a similarity comparison between the embedding representation of the query and a set of embedding representations of historical document sections stored in a historical document repository (fig. 1C, items 172 and 174, par. 51, “query representations correspond to the embedding representations of the asset index library), wherein each embedding representation of a historical document section satisfies a token limit for a target large language model and each embedding representation of a historical document section represents components of a single historical document section within a historical document (figs. 1C and 2, pars. 59-60, each embeddings and asset representations are trained to encode generic knowledge of semantic concepts wherein query representation models 176 are trained to map concepts and tokens in input queries to concepts of asset training data, the asset training data and index library is equated to historical document and historical document section);
selecting, by the analysis circuitry and based on the similarity comparison, a relevant embedding representation of a historical document section stored in the historical document repository for the query (fig. 1D illustrates the training asset representation models and query representation models, see pars. 51, 52 and 59, ML models used in conjunction with plurality of representation models to perform searches of multimodal queries based on similarity);
querying, by the analysis circuitry, the target large language model using the embedding representation of the query and the relevant embedding representation of the historical document section (fig. 6, items 620-625, pars. 71-72, query and compare multimodal data to multimodal assets, by using different models as cited above); and
providing, by the communications hardware, a query response to the user device (fig. 6, item 630, par. 73, search result is provided).
2. The method of claim 1, further comprising: Krishnan discloses receiving, by the communications hardware, a historical document, partitioning, by the analysis circuitry, the historical document into one or more historical document sections, performing, by the analysis circuitry, a document embedding routine for each historical document section of the one or more historical document sections, the document embedding routine comprising: tokenizing, by the analysis circuitry, a historical document section into a plurality of historical document section tokens, and generating, by the analysis circuitry and based on the plurality of historical document section tokens, an embedding representation of the historical document section and storing, by the analysis circuitry, the embedding representation for each historical document section in the historical document repository (fig. 1A, item 122, par. 37, repository is representative of databases relating to training models, asset libraries and vectorized representations of assets and stored. See pars. 45 and 51, multimodal assets comprise tensor generation including vector representations for each of the elements of the asset and provides multilinear relationship between the vector representations in the tensor. Each row/section may represent different modality of each asset such as text segments and image segments).
3. The method of claim 2, further comprising: Krishnan discloses determining, by the analysis circuitry, whether a number of historical document section tokens for the historical document section would exceed a maximum token limit and in response to determining the maximum token limit would be exceeded, partitioning, by the analysis circuitry, the historical document section into a plurality of historical document subsections, wherein each of the plurality of historical document subsections may be tokenized into a plurality of historical document section tokens (pars. 50 and 59, size restrictions and specific types of data is analyzed/partitioned with the concepts and tokens in input queries and concepts of asset data).
4. The method of claim 2, further comprising: Krishnan discloses detecting, by the analysis circuitry, a first modality type for content within a historical document section; detecting, by the analysis circuitry, a second modality type for the content within the historical document section, wherein the second modality type is different than the first modality type; and in response to detecting that the second modality type is different than the first modality type, partitioning, by the analysis circuitry, the historical document section into a plurality of historical document subsections based on a corresponding modality type, wherein each of the plurality of historical document subsections may be tokenized into a plurality of historical document section tokens (see pars. 45 and 51, multimodal assets comprise tensor generation including vector representations for each of the elements of the asset and provides multilinear relationship between the vector representations in the tensor. Each row/section may represent different modality of each asset such as text segments and image segments).
5. The method of claim 1, further comprising: Krishnan discloses determining, by the analysis circuitry, whether the query requests an evaluation for an input document; in response to determining that the query requests the evaluation for the input document, partitioning, by the analysis circuitry, the input document into a plurality of target document sections, tokenizing, by the analysis circuitry, each target document section into a plurality of target document section tokens generating, by the analysis circuitry and based on the plurality of target document section tokens, an embedding representation of each target document section and querying, by the analysis circuitry, the target large language model using the embedding representation for each target document section and the relevant embedding representation of the historical document section (fig. 1C, pars. 49 and 51, text query is submitted as a search query and a multimodal document, the query is converted to respective embeddings based on content type, i.e., textual and image and (figs. 1C and 2, pars. 59-60, each embeddings and asset representations are trained to encode generic knowledge of semantic concepts wherein query representation models 176 are trained to map concepts and tokens in input queries to concepts of asset training data, the asset training data and index library is equated to historical document and historical document section), wherein the query response further comprises an evaluation assessment and the evaluation assessment comprises an evaluation status for each target document section (pars. 35 and 50, user query comprises a plurality of requirements and thresholds, such as image must match at least 95%, this requires the models to retrieve the assets that comply to the query requirements).
6. The method of claim 5, Krishnan discloses wherein the evaluation assessment comprises an evaluation reason for the evaluation status (pars. 35 and 50, user query comprises a plurality of requirements and thresholds, such as image must match at least 95%, this requires the models to retrieve the assets that comply to the query requirements, see also par. 56, the evaluation reason is meeting additional data 138 criteria such as specified asset type).
7. The method of claim 5, Krishnan discloses wherein evaluation assessment further comprises a recommendation for a target document section (pars. 39 and 56, content recommendation and meeting criteria specified).
8. The method of claim 5, Krishnan discloses further comprising performing, by automation circuitry, a proactive action in an instance in which the evaluation assessment comprises at least one evaluation status indicative that a target document section is non-compliant (pars. 35 and 50, user query comprises a plurality of requirements and thresholds, such as image must match at least 95%, this requires the models to retrieve the assets that comply to the query requirements, see also par. 56, the evaluation reason is meeting additional data 138 criteria such as specified asset type, result may indicate no result meets the requirement).
9. The method of claim 8, Krishnan discloses wherein performing the proactive action further comprises automatically updating, by the automation circuitry, the input document to (i) modify current language or (ii) insert new language (pars. 24, 29 and 39, system automatically provides recommendations for inserting content into a document, interactively edit, and par. 50, may insert certain types of visual assets into decuments).
10. The method of claim 5, Krishnan discloses further comprising: detecting, by the analysis circuitry, a first modality type for content within a target document section, detecting, by the analysis circuitry, a second modality type for the content within the target document section, wherein the second modality type is different than the first modality type, and in response to detecting that the second modality type is different than the first modality type, partitioning, by the analysis circuitry, the target document section into a plurality of target document subsections based on a corresponding modality type, wherein each of the plurality of target document subsections may be tokenized into a plurality of target document section tokens (see pars. 45 and 51, multimodal assets comprise tensor generation including vector representations for each of the elements of the asset and provides multilinear relationship between the vector representations in the tensor. Each row/section may represent different modality of each asset such as text segments and image segments for query and asset matching).
11. The method of claim 5, Krishnan discloses further comprising: determining, by the analysis circuitry, whether a number of target document section tokens for a target document section would exceed a maximum token limit and in response to determining the maximum token limit would be exceeded, partitioning, by the analysis circuitry, the target document section into a plurality of target document subsections, wherein each of the plurality of target document subsections may be tokenized into a plurality of target document section tokens (pars. 50 and 59, size restrictions and specific types of data is analyzed/partitioned with the concepts and tokens in input queries and concepts of asset data).
12. The method of claim 1, further comprising: Krishnan discloses performing, by training circuitry, a training routine, the training routine comprising: initializing, by the training circuitry, a base large language model, and adjusting, by the training circuitry, the base large language model based on a domain-specific training data set (pars. 31, 32 and 57, labeling data may be initially used to train LLM, supplemental data and training may be added to fine tune, labeling implies domain-specific training).
13. The method of claim 12, further comprising; Krishnan discloses receiving, by the communications hardware, annotated authentic domain-specific documents, wherein: each annotated authentic domain-specific document includes (i) a corresponding authentic domain-specific document, (ii) a training query prompt, (iii) and an expected response to the training query prompt, the domain-specific training data set comprises the annotated authentic domain- specific documents (pars. 39, 56 and 70, search results comprise documents responsive to query prompt and content recommendation such as annotations).
14. The method of claim 1, Krishnan discloses wherein the historical document is at least one of a model development document and a model validation document (pars. 32-33, different types of assets, the training data may be obtained from a repository or device generated thus model development, data is validated).
Regarding apparatus claims 16-19, they comprise substantially the same subject matter as rejected method claims 2-14, above, and are therefore rejected on the merits.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure in the field of embedded searching:
USPN. 20250111157 USPN. 20250156634 Abstract: multi LLM embedded searching
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MARCIN R FILIPCZYK whose telephone number is (571)272-4019. The examiner can normally be reached M-F 7-4 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kavita Stanley can be reached at 571-272-8352. 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.
January 9, 2026
/MARCIN R FILIPCZYK/Primary Examiner, Art Unit 2153