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 .
Specification
The disclosure is objected to because of the following informalities:
Paragraph 00022 Line 9 recites “Fror example”
Paragraph 00023 Line 2 recites “QT-FT” which should be corrected to “QA-FT”
Appropriate correction is required.
Drawings
The drawings are objected to because:
Fig. 3D contains the phrase “Continous-FT” three times. This should be corrected to “Continuous-FT”.
Fig. 4 contains the phrase “Generale query-response…” in part 404. This should be corrected to “Generate…”
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 § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 13 recites the limitation "the context" in Line 1. There is insufficient antecedent basis for this limitation in the claim.
Claim Rejections - 35 USC § 102
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) 1-2, 4, 5, 8 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Lee (LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancement).
Regarding Claim 1, Lee teaches a method comprising:
providing a data sample(new target domain by using a small seed dataset D) comprising a query(Question[Figure 1], “Please select the correct holding statement from the options below”[Figure B.2]) and a context(In practice, we also use in-context learning with few-shot prompting [B.4 Prompting details], Example of domain context: “Context: Drapeau’s cohorts, the cohort would…”[Figure B.2]) of the query to a first large language model (LLM)(M^i student [Algorithm 1], Student LLM fine-tuned on the initial seed data[Abstract]) using an LLM prompt(Example of prompt: Figure B.2), wherein the LLM prompt causes the first LLM to generate a response to the query(Example of response to the query: D [Figure B.2]);
determining, based on the providing, the response(M_student on the provided target data
D0, and we evaluate its performance (lines 4-5 of Algorithm 1)[3.1 LLM2LLM]) and an annotation to the response that indicates a relevancy of the response to the query in the context of the query;(the teacher model has access to how the student model performs at the nth step (e.g., correct/incorrect labels [3 Methodology])
generating, based at least on the data sample, the annotation, and a plurality of other annotated data samples(teacher model Mteacher to generate synthetic data from the data points that the model got incorrect during training in order to target these deficiencies in the student model [3.1 LLM2LLM], Figure B.9), a training data set(A^i ←Generate(M_teacher, W^i) usable to train a second LLM(M^0 student, [Algorithm 1]) using a fine-tuning technique(generate new training data points using the teacher model [Algorithm 1], Finetune(M^0 student, Di)[Algorithm 1]);
performing a data sampling(Once the teacher model generates the synthetic data, we need to apply simple filtering of the output for quality insurance. Like in previous work (Wang et al., 2023; Taori et al., 2023), we use regex filters to ensure that the basic format of the output is aligned with our expectations. We also use a ROUGE (Lin, 2004) filter in order to enforce that the augmented data points are sufficiently different from previous samples.[B.3 Filtering Generated Dataset]).) and a data augmentation(teacher model has enough reasoning capability to produce conceptually similar but semantically different examples[3. Methodology], 7: Ai ←Generate(Mteacher, Wi) ▷ Augment using teacher[Algorithm 1]) on the training data set;
updating the training data set based on the data sampling and the data augmentation performed(D^i+1 <- D^i +A^i [Algorithm 1], wherein D represents the training dataset); and
training the second LLM(M^0 student, [Algorithm 1]) using the updated training data set(D^i+1) and the fine-tuning technique, wherein the fine-tuning technique is applied to the second LLM(Finetune(M^0 student, D^i)[Algorithm 1]) when training based on a number of data samples in the updated training data set(The fine-tuning step trains a small student model using seed data and previously generated LLM2LLM data[B.1 Fine-tuning], Interpretation: at each iteration the fine-tuning performed with different amounts of data).
Regarding Claim 2, Lee teaches the context is associated with a domain of data(a medical dataset with specific terminology, or a private database with specific characteristics[3. Methodology]) in which the query is requested for the response from a chatbot, and wherein the chatbot provides domain-specific responses based on domain knowledge(Example of response: D [Figure B.2]) associated with a plurality of domain documents(Examples of domain documents corresponding to the domain of data: Halleck v. Berlinger, 427 F. Supp. 1225, 1241 (D.D.C. 1977) , Mississippi Comm’n on Judicial Performance v. Boland, 975 So.2d 882, 891-92 (Miss. 2008) [Figure B.2]) corresponding to the domain of data.
Regarding Claim 4, Lee teaches the determining the response and the annotation comprises: receiving the annotation, wherein, when the relevancy indicates that the response is not relevant to the query(Incorrect answers from the training data are used as inputs to generate extra samples with similar styles to the teacher model[Figure 1 Description]), the annotation further includes an amendment to the response; and creating the data sample(Interpreted as part of the prompt given to the teacher model to generate the training data set) including an information set indicating the prompt(example of a prompt: Figure B.8), the query(Question[Figure 1], “Please select the correct holding statement from the options below”[Figure B.9]), the context(Example of domain context: Context: Drapeau’s cohorts…[Figure B.9]), one of the responses(Example of response: Answer: A [Figure B.9]) or the amendment to the response(Example of the amendment in response to wrong answer: The following is a multiple choice question about the holding statements of a judicial decision that the user got wrong including the correct answer from the answer sheet:[Figure B.9]), and a label corresponding to the relevancy from the annotation(the teacher model has access to how the student model performs at the nth step (e.g., correct/incorrect labels [3 Methodology]).
Regarding Claim 5, Lee teaches the fine-tuning technique includes at least one of a question-answering (QA) fine-tuning(Lee does not explicitly teach QA-FT however the dataset used for fine-tuning utilizes QA pairs), a RAG fine-tuning, or a continuous fine-tuning, and wherein the RAG fine-tuning is utilizable with an outcome-based training and a process-based training for the training the second LLM.
Regarding Claim 8, Lee teaches the data augmentation comprises at least one of a query augmentation(1. Question: Betty is saving money for a new wallet [Figure B.7], Interpretation: teacher model is augmenting the training dataset by adding new queries), a response augmentation, or a context augmentation.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 3, 7 are rejected under 35 U.S.C. 103 as being unpatentable over Lee (LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancement) in view of Zhang(RAFT: Adapting Language Model to Domain Specific RAG [Reference disclosed in IDS]).
Regarding Claim 3, Lee does not teach generating, using a retrieval augmented generation (RAG) operation of the first LLM, the response based on the query and the plurality of domain documents.
However, Zhang teaches generating, using a retrieval augmented generation (RAG)( RAG based in-context learning) operation of the LLM, the response based on the query and the plurality of domain documents(RAG based methods allow the LLM to reference the documents when answering questions. Model can use External Docs at Test [Figure 1]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the method of Lee with the RAG of Zhang because it would allow the LLM to reference the documents when answering questions(Zhang).
Regarding Claim 7, Lee as above in Claim 1, teaches the outcome-based training uses a requested response to each query from the training data set during a fine-tuning loss computation(Finetune(M^0 student, D^i)[Algorithm 1], Interpretation: fine-tuning uses the response from M^0 student to the query, where fine-tuning uses a loss computation to adjust model parameters).
Lee does not teach wherein the process-based training uses a chain-of-thought (CoT) reasoning response during the fine-tuning loss computation.
However, Zhang teaches the outcome-based training uses a requested response to each query from the training data set during a fine-tuning loss computation, and wherein the process-based training uses a chain-of-thought (CoT) reasoning response(we prepare the training data such that each data point contains a question (Q), a set of documents (Dk), and a corresponding Chain-of-though style answer (A∗) generated from one of the document (D∗)[3. RAFT]) during the fine-tuning loss computation.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the method of Lee with the training methodology of Zhang because it would enhance training quality(Zhang).
Claim 6 is rejected under 35 U.S.C. 103 as being unpatentable over Lee (LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancement) in view of Zhang(RAFT: Adapting Language Model to Domain Specific RAG [Reference disclosed in IDS]) and further in view of Navon(US PGPub 20250165750).
Regarding Claim 6, Lee as above in Claim 1, teaches performing continuous fine-tuning(Another key decision for LLM2LLM is whether to continue fine-tuning from the last iteration’s checkpoint (i.e. continuous fine-tuning)[4.5.3 From-scratch Fine-tuning vs Continuous Fine-tuning]) during a plurality of iterations(We
can observe a rapid increase in test accuracy in the first few iterations of LLM2LLM, especially in lower-data regimes.[4.2 Main results]) of the training the second LLM.
Lee does not teach when the number of data samples is less than or equal to a threshold number, the training the second LLM uses the QA fine-tuning before the RAG fine-tuning, or wherein, when the number of data samples is greater than or equal to the threshold number, the training the second LLM uses RAG fine-tuning.
However, Zhang in view of Navon teaches when the number of data samples is greater than or equal to the threshold number(FT data identifier 12 can consider characteristics of the data, such as the application providing the data, a size of the data (e.g., data smaller than a threshold size… [Navon 0022]), the training the second LLM uses RAG fine-tuning(Retrieval Augmented Fine Tuning (RAFT), a training recipe which improves the model’s ability to answer questions in "open-book" in-domain settings[Zhang Abstract]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the method of Lee with the RAG-FT of Zhang and the dataset threshold of Navon because it would improve the model’s ability to answer questions in "open-book" in-domain settings(Zhang) and because it would better streamline fine-tuning of LLMs(Navon).
Claims 9,13-14,17 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma(Retrieval Augmented Generation for Domain-specific Question and Answering) in view of Rennie(US PGPub 20240265041 [Reference disclosed in IDS]).
Regarding Claim 9, Sharma teaches a method comprising:
providing a plurality of domain documents with document metadata to a first large language model (LLM) using an LLM prompt(We prepend a couple of sample documents as well as generated QA pairs with each call to our LLM, in order to guide the generation style and quality of the question.[3.4.1 QA Generation Module]), wherein the LLM prompt causes the first LLM to generate a plurality of query-response pairs(we generate multiple QA pairs from each document using a QA generation module powered by a Large Language model [3.3 Retrieval Index Creation and Database]) based on the document metadata(Helpx Documents (title, description) – We take the title and descriptions of helpx documents and embed them.[3.3 Retrieval Index Creation and Database]);
determining, based on the providing, the plurality of query-response pairs each corresponding to a source document of the plurality of domain documents(we generate multiple QA pairs from each document using a QA generation module powered by a Large Language model [3.3 Retrieval Index Creation and Database]);
generating, based at least on the plurality of query-response pairs, a training data set(derived datasets [Preprocessing for Building the Database]) usable to train a second LLM using a fine-tuning technique;
performing a data sampling(named entity removal module to provide privacy safeguards for the dataset[Preprocessing for Building the Database]) and a data augmentation(product-intent extraction model(sharma2024semantic,) that maps all input texts to 1 or more Adobe products. We then pass this information to the retriever, allowing for better relevancy in the retrieved documents [Query Augmentation via Product Identification]) on the training data set;
updating the training data set based on the data sampling and the data augmentation performed; and
training the second LLM using the updated training data set(To train the LLM, we use grounded document, d+, negative document d−, and question-answer pairs, (q,a).[LLM Finetuning]) and the fine-tuning technique, wherein the fine-tuning technique is applied to the second LLM when training based on a number of data samples in the updated training data set(Interpretation: Fine-tuning is performed using a number of question-answer pairs).
Sharma does not teach generating a plurality of additional queries from at least one of the source documents or a plurality of top-n documents retrieved for each query of the plurality of query-response pairs.
However, Rennie teaches generating a plurality of additional queries from at least one of the source documents or a plurality of top-n documents retrieved for each query of the plurality of query-response pairs(modifying a source question from the question dataset to generate one or more augmented questions with equivalent semantic meanings as that of the source question[Abstract]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine the method of Sharma with the question generation of Rennie because it would improve the quality of searching operations(Rennie).
Regarding Claim 13, Sharma as above in Claim 9, teaches the context is associated with a domain of data in which the query is requested for the response from a chatbot, and wherein the chatbot provides domain-specific responses based on domain knowledge associated with a plurality of domain documents corresponding to the domain of data (We prepend a couple of sample documents as well as generated QA pairs with each call to our LLM, in order to guide the generation style and quality of the question.[3.4.1 QA Generation Module]).
Regarding Claim 14, Sharma as above in claim 9, teaches the fine-tuning technique includes at least one of a question-answering (QA) fine-tuning(To train the LLM, we use grounded document, d+, negative document d−, and question-answer pairs, (q,a).[3.7 Fine-tuning]), a RAG fine-tuning, or a continuous fine-tuning, and wherein the RAG fine-tuning is utilizable with an outcome-based training and a process-based training for the training the second LLM.
Regarding Claim 17, Sharma as above in Claim 9, teaches the data augmentation comprises at least one of a query augmentation(product-intent extraction model(sharma2024semantic,) that maps all input texts to 1 or more Adobe products. We then pass this information to the retriever, allowing for better relevancy in the retrieved documents [Query Augmentation via Product Identification]), a response augmentation, or a context augmentation.
Claim(s) 10-11 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma(Retrieval Augmented Generation for Domain-specific Question and Answering) in view of Rennie(US PGPub 20240265041 [Reference disclosed in IDS]) as applied to claim 9 above, and further in view of Yan(Corrective Retrieval Augmented Generation).
Regarding Claim 10, neither Sharma nor Rennie teach prior to generating the training data set, the method further comprises: identifying the plurality of top-n documents retrieved for each query in the plurality of query-response pairs; and determining whether the source document is among the plurality of top-n documents, wherein, when the source document is among the plurality of top-n documents, a corresponding one of the plurality of query-response pairs and one or more corresponding queries of the plurality of additional queries are annotated with a positive response annotation, or wherein, when the source document is not among the plurality of top-n documents, a corresponding one of the plurality of query-response pairs and one or more corresponding queries of the plurality of additional queries are annotated with a negative response annotation.
However, Yan teaches prior to generating the training data set, the method further comprises: identifying the plurality of top-n documents retrieved for each query in the plurality of query-response pairs(Retrieved Documents [Figure 2]); and determining whether the source document is among the plurality of top-n documents(A retrieval evaluator is constructed to evaluate the relevance of the retrieved documents to the input, and estimate a confidence degree based on which different knowledge retrieval actions of {Correct, Incorrect, Ambiguous} can be triggered[Figure 2]), wherein, when the source document is not among the plurality of top-n documents, a corresponding one of the plurality of query-response pairs and one or more corresponding queries of the plurality of additional queries are annotated with a negative response annotation(Incorrect Besides, a retrieval is assumed
Incorrect when the confidence scores of all retrieved documents are below the lower threshold. This indicates that all retrieved documents are considered irrelevant, which are unhelpful for generation.[4.3 Action Trigger]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine the method of Sharma modified by Rennie with the retrieval analysis of Yan because it would reduce the amount of irrelevant information(Abstract, Yan).
Regarding Claim 11, as Claim 10 was rejected under the assumption that the source document is not among the plurality of top-n documents, Claim 11 would not happen. However, if applicant amends the claim to ensure that the source document is among the plurality of top-n documents, this claim may become allowable if written in independent form.
Claim(s) 12 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma(Retrieval Augmented Generation for Domain-specific Question and Answering) in view of Rennie and Yan as applied to claim 10 above, and further in view of Lee (LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancement).
Regarding Claim 12, Rennie as above in Claim 9, teaches the first LLM generates the plurality of additional queries from each query of the plurality of query-response pairs(a question (original or variant) is provided to the generative LLM engine to produce a linguistically different question than that retrieved from the database 302, but with a semantically similar meaning.[0177]).
Neither Sharma nor Yan teach that the additional queries are generated based on spelling deviations and grammatical deviations.
However, Lee teaches that additional queries are generated based on spelling deviations and grammatical deviations(For natural language processing (NLP) tasks, one can use approaches such as synonym replacement, character replacement (e.g., by intentionally introducing spelling errors), random swapping, and back translation[1. Introduction]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine the method of Sharma modified by Rennie with the query augmentation of Lee because it would effectively expand the training dataset(Lee).
Claim(s) 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma(Retrieval Augmented Generation for Domain-specific Question and Answering) in view of Rennie(US PGPub 20240265041 [Reference disclosed in IDS]) as applied to claim 9 above, and Zhang(RAFT: Adapting Language Model to Domain Specific RAG [Reference disclosed in IDS]) and further in view of Navon(US PGPub 20250165750).
Arguments analogous to Claim 6 and 7 respectively are applicable.
Claim(s) 18 is rejected under 35 U.S.C. 103 as being unpatentable over Lee (LLM2LLM: Boosting LLMs with Novel Iterative Data Enhancement) in view of Hurst(US PGPub 20250307286).
Regarding Claim 18, Lee teaches generating, based a plurality of query-response pairs and annotations generated by a first large language model (LLM), a training data set usable to train a second LLM using a fine-tuning technique(teacher model Mteacher to generate synthetic data from the data points that the model got incorrect during training in order to target these deficiencies in the student model [3.1 LLM2LLM], Figure B.9); performing a data sampling(Once the teacher model generates the synthetic data, we need to apply simple filtering of the output for quality insurance. Like in previous work (Wang et al., 2023; Taori et al., 2023), we use regex filters to ensure that the basic format of the output is aligned with our expectations. We also use a ROUGE (Lin, 2004) filter in order to enforce that the augmented data points are sufficiently different from previous samples.[B.3 Filtering Generated Dataset]).) and a data augmentation(teacher model has enough reasoning capability to produce conceptually similar but semantically different examples[3. Methodology]) on the training data set;
updating the training data set based on the data sampling and the data augmentation performed; and training the second LLM using the updated training data set and the fine-tuning technique, wherein the fine-tuning technique is applied to the second LLM when training based on a number of data samples in the updated training data set((The fine-tuning step trains a small student model using seed data and previously generated LLM2LLM data[B.1 Fine-tuning], Interpretation: at each iteration the fine-tuning performed with different amounts of data).).
Lee does not teach a non-transitory memory; and one or more hardware processors coupled to the non-transitory memory and configured to read instructions from the non-transitory memory to cause the service provider system to perform operations.
However, Hurst teaches non-transitory memory(computing device 600 and can include both volatile and nonvolatile storage media and removable and non-removable storage media. Tangible processor-readable storage media excludes intangible, transitory communications signals [0089]); and one or more hardware processors coupled to the non-transitory memory and configured to read instructions from the non-transitory memory to cause the service provider system to perform operations(The computing device 600 includes one or more hardware processor(s) 602, In some implementations, the computing device 600 includes and/or is communicatively coupled to storage 620.[0084]).
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine the method of Lee with the hardware of Hurst because it would allow for the method to be implemented on hardware(Hurst).
Regarding Claim 19, Lee modified by Hurst as above in Claim 18, teaches the annotations are generated by the first LLM using retrieval augmented generation (RAG)( a large language model (LLM) can base responses to RAG-assistant-generated prompts not only on the verbatim content of the source documents in the index, but also provide annotations[Hurst 0012]) and wherein, subsequent to generating the annotations, a data sampling(Once the teacher model generates the synthetic data, we need to apply simple filtering of the output for quality insurance. Like in previous work (Wang et al., 2023; Taori et al., 2023), we use regex filters to ensure that the basic format of the output is aligned with our expectations. We also use a ROUGE (Lin, 2004) filter in order to enforce that the augmented data points are sufficiently different from previous samples.[Lee B.3 Filtering Generated Dataset]).) and a data augmentation(teacher model has enough reasoning capability to produce conceptually similar but semantically different examples[Lee 3. Methodology]) is performed to generate additional query-response pairs added to the plurality of query-response pairs prior to the generating the training data set(7: Ai ←Generate(Mteacher, Wi) ▷ Augment using teacher[Lee Algorithm 1]).
Regarding Claim 20, Lee modified by Hurst as above in Claim 18, teaches the annotations are generated by the first LLM using domain documents based on a source document(an annotation that cites (1) a source document that the synthetic chunk 207 was derived from[0046] Interpretation: synthetic chunk maps to domain document) for each of the plurality of query-response pairs(teacher model has access to how the student model performs at the nth step[Lee 3. Methodology]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. US PGPub 20250322242. This publication discloses a method and system for training an LLM to generate domain-specific fine-tuning training data and then performs the fine-tuning on a separate LLM.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ARJUN R SWAMY whose telephone number is (571)272-9763. The examiner can normally be reached Mon, Tue, Thur, Fri 8-5.
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, Hai Phan can be reached at (571) 272-6338. 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.
/ARJUN SWAMY/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654
Prosecution Insights