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 .
DETAILED ACTION
1. This Office Action is in response to the Amendment filed on February 25, 2026, which paper has been placed of record in the file.
2. Claims 1-20 are pending in this application.
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Claim Rejections - 35 USC § 103
3. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
4. Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Pyle (US 2025/0007794) in view of Qin et al. (hereinafter Qin, US 2024/0346256).
Regarding to claim 1, Pyle discloses a non-transitory computer-readable medium having instructions stored to generate decision recommendations to optimize asset allocation in an enterprise, wherein the instructions, when executed by a processor, cause the processor to:
obtain enterprise data information from more than one source (para [0077], The peer ranking module 140 may include computer-executable instructions for collecting data based on activities of the clouds in the could environment ENV1. For example, the peer ranking module 140 may receive information from the hyperscaler module 130 and process the information to determine the purposes for which the cloud environment ENV1 is being used),
wherein the data information is associated with users, asset classes and costs (para [0065], The analytics may include statistics regarding any use of the respective clouds, such as network bandwidth usage, CPU load, data types, industry affiliations, etc.; para [0055], The present techniques may provide the ability to procure cloud licensure/subscriptions and to have multi-cloud financial snapshots, including unified cloud invoicing, budgeting, cost anomaly detection, and ongoing helpdesk support),
wherein the processor detects undiscovered existing users and assets (para [0065], the application may include a machine learning model trained by the machine learning training module 134 that is specifically trained to detect patterns regarding certain types of information (e.g., information subject to HIPAA protection));
process the data information to identify valid users and asset classes, anomalies, and relationships between usage and costs (para [0054], the present techniques may provide a single consolidated cloud invoice that allows cloud service consumers to easily reconcile and charge to the associated department, or project. The present techniques may also include cloud cost anomaly detection aspects that immediately alert customers and IT to an unusual spike in activity, advantageously enabling the consumers to immediately rectify such issues, without realizing a dramatic fiscal impact to the business);
generate a mapping of the valid users and valid asset classes to define a relationship and associated trend between an enterprise asset, an enterprise user, and enterprise costs (para [0089], In supervised machine learning, a machine learning program operating on a server, computing device, or otherwise processor(s), may be provided with example inputs (e.g., “features”) and their associated, or observed, outputs (e.g., “labels”) in order for the machine learning program or algorithm to determine or discover rules, relationships, patterns, or otherwise machine learning “models” that map such inputs (e.g., “features”) to the outputs (e.g., labels), for example, by determining and/or assigning weights or other metrics to the model across its various feature categories. Such rules, relationships, or otherwise models may then be provided subsequent inputs in order for the model, executing on the server, computing device, or otherwise processor(s), to predict, based on the discovered rules, relationships, or model, an expected output);
derive a usage information on the enterprise asset by the enterprise user based on the relationship and associated trend between the enterprise asset, enterprise user, and enterprise costs (para [0089], Machine learning may involve identifying and recognizing patterns in existing data (such as data risk issues, data quality issues, data sensitivity, resource usage, industry affiliation, data type, etc.) in order to facilitate making predictions, classifications, and/or identifications for subsequent data);
provide the usage information on the enterprise asset by the enterprise user to an artificial intelligence (AI) system for data analysis and pattern recognition (para [0090], For example, the ML training module 134 may analyze labeled historical data at an input layer of a model having a networked layer architecture (e.g., an artificial neural network, a convolutional neural network, a deep neural network, etc.) to generate ML models. The training data may be, for example, data collected from hyperscalers of one or more customers (e.g., those in similar industries). The historical data may include labels. During training, the labeled data may be propagated through one or more connected deep layers of the ML model to establish weights of one or more nodes, or neurons, of the respective layers),
wherein the AI system includes one or more large language model module configured to perform an analysis of the usage information on the enterprise asset by the enterprise user to optimize asset allocation in an enterprise (para [0068], In some aspects, the machine learning training model 134 may fine-tune an existing GPT model. The existing model may be, for example, an open source large language model (LLM) (e.g., Large Language Model Meta AI (LLaMA)) or another small-parameter or large-parameter LLM. The fine-tuning may include grounding a pre-trained model by providing additional training examples using a zero-shot or few-shot learning strategy);
wherein the large language model module utilizes a natural language processing interface (para [0088], The machine learning programs or algorithms may also include natural language processing, semantic analysis, automatic reasoning, regression analysis, support vector machine (SVM) analysis, decision tree analysis, random forest analysis, K-Nearest neighbor analysis, naïve Bayes analysis, clustering, reinforcement learning, and/or other machine learning algorithms and/or techniques);
analyze the recommendation based on whether the recommendations have materialized in the enterprise (para [0074], The recommendation engine 136 may include analyzing the cloud computing usage of the company/client to determine one or more patterns. In some cases, one or more machine learning models may be used to determine such patterns. For example, a trends module (not depicted) may determine that the company keeps 50% (or less) of cloud computing CPU utilization across a plurality of instances and/or across cloud compute providers. In that case, the recommendation engine may infer that the low utilization is intentional, and may adjust the set of weights corresponding to that customer/client, to customize the recommendations given to that client),
wherein the enterprise performs an action based on whether the recommendation has materialized (para [0139], the method 800 may show some or all of the actions 524 as opportunities for improvement, when the processing of data from the one or more hyperscaler instances indicates that the status of the action is False. The recommendations include recommendations that correspond to a plurality of different cloud computing providers. This represents a solution to a problem identified in conventional cloud management practices, which is that users are required to access multiple different cloud management interfaces, all of which have different management interfaces, to attempt to correct issues in an ad hoc and disorganized fashion); and
use a result of the optimizing of the asset allocation to train the large language model module (para [0141], the machine learning training module 134 may train a model to output recommendations regarding the toggling of configuration parameters to avoid the creation of such large log files. In some aspects, the machine learning training module 134 may train a supervised machine learning model using labeled training data. The labeled training data may include a set of log messages related to security vulnerabilities or outdated software. The machine learning training module 134 may train the supervised model to identify de novo messages related to such issues within the user's cloud environment, and to generate recommendations (actions) and cloud maturity weights for fixing such issues).
Pyle does not disclose, however, Qin discloses:
wherein the large language model module is enhanced using retrieval- augmented generation (RAG) that combines generative capabilities with a retrieval mechanism to retrieve relevant information from a dataset and generate contextually relevant responses through text processing, thereby improving the large language model module's ability to provide insight and recommendation for asset allocation (para [0018], Embodiments are disclosed herein that improve the scope and accuracy of responses generated by an LLM. For instance, in embodiments, an LLM may be augmented with augmentation information (e.g., domain-specific information; entity-specific information; product-specific information; recent information unavailable at generation of the large language model; or information changed after generation of the large language model). A retrieval augmented generation (RAG) approach is disclosed herein that adds an information retrieval component to create augmented prompts to feed into the generative language model for generating the final answer/prediction. RAG is a general-purpose fine-tuning which combines pre-trained parametric and non-parametric memory for language generation. The pre-trained LLM such as GPT3 contains parametric memory. The non-parametric memory is a vector dictionary. A knowledge base is built for domain-specific content. This is accomplished with “dense vector embeddings”, which are numerical representations of the meaning behind content/sentences);
wherein the large language model module processes text input to perform intent detection and entity recognition, and creates the intent into a schema mapping to retrieve and detect assets within the enterprise, and constructs a query to retrieve requested data based on the schema mapping (para [0019], a query may be used to retrieve pieces of augmentation information that may be included in a prompt to the LLM. For instance, a query string may be encoded into a first feature vector that is compared to a plurality of second feature vectors to determine a subset of the second feature vectors that satisfy a predetermined condition (e.g., threshold similarity). Augmentation information corresponding to the determined subset of second feature vectors may be retrieved and included in an augmented prompt to the LLM. In embodiments, the augmented prompt may include the original query, contextual information for answering the query, the retrieved augmentation information, and/or a request to answer the original query based on the contextual information and/or the retrieved augmentation information. When presented with the retrieved augmentation information, the LLM prioritizes the retrieved augmentation information over the information present in its training data when generating a response to the query. Queries generated based on the augmented information have the benefit of generally more focused and accurate).
Therefore, it would have been obvious to one with ordinary skill in the art before the effective filing date of the claimed invention to modify the Pyle’s to incorporate the features taught by Qin above, for the purpose of improving the accuracy and contextual relevance of generated outputs when applying the LLM utilizing with a retrieval augmented generation (RAG). Since Pyle discloses performing the analysis using the LLM, Qin discloses using the LLM with the RAG, as described above, therefore, one of ordinary skill in the art would have recognized that the combination of Pyle and Qin would have yield predictable results in performing the analysis using the LLM.
Regarding to claim 2, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the data information comprises information from a cloud (para [0081], The environment 160 may include a billing engine that allows for proration of cloud/SaaS costs with the ability to bundle professional and managed services that can be invoiced on a monthly, quarterly, or annual basis. The billing engine may generate unified cloud invoices for customers and, on a monthly basis ingest billing information (e.g., CSV file into its accounting system)).
Regarding to claim 3, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the data information comprises information from an enterprise premises data source (para [0124], For example, the IT services company may process historical billing data to determine that enabling an autoscaler in the data explorer of value 520-F has resulted in a proportionally large cost reduction. The machine learning model may be trained to output a large or smaller weight value based on this information).
Regarding to claim 4, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the data information comprises information from a cloud and enterprise premises data sources (para [0055], the present techniques may provide cloud computing sales, management, reporting, governance, and optimization; all in one platform. The present techniques may provide the ability to procure cloud licensure/subscriptions and to have multi-cloud financial snapshots, including unified cloud invoicing, budgeting, cost anomaly detection, and ongoing helpdesk support).
Regarding to claim 5, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the processor further derives the usage information on the enterprise asset by the enterprise user by evaluating demographics, conducting cost analysis, addressing security and compliance issues, and providing detailed usage data history to determine the usage and location of the assets (para [0074], The recommendation engine 136 may include analyzing the cloud computing usage of the company/client to determine one or more patterns. In some cases, one or more machine learning models may be used to determine such patterns. For example, a trends module (not depicted) may determine that the company keeps 50% (or less) of cloud computing CPU utilization across a plurality of instances and/or across cloud compute providers).
Regarding to claim 6, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein users can provide feedback to train the large language model module (para [0069], Differing sets of training examples may be used to fine-tune one or more GPTs/LLMs in different ways. For example, cloud computing data may be stratified by industry, as discussed herein. Data for one industry (e.g., healthcare) may be selected and used to fine tune an LLM. This fine-tuned model may be called via an API and provided with information about a new customer).
Regarding to claim 7, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the processor can automatically incorporate new data information (para [0095], The model operation module may apply new data that the trained model has not previously analyzed to the trained model. For example, the model operation module may load a serialized model, deserialize the model, and load the model into memory).
Regarding to claim 8, Pyle discloses the non-transitory computer-readable medium of claim 1, further comprising conversational agent interface configured to allow for a user to converse with an agent using the large language model module and a natural language processing (NLP) (para [0091], In the present techniques, unsupervised learning may be used, inter alia, for natural language processing purposes and to identify scored features that can be grouped to make unsupervised decisions (e.g., numerical k-means)).
Regarding to claim 9, Pyle discloses the non-transitory computer-readable medium of claim 8, wherein the NLP is configured to dynamically create charts and graphs based on user requests (para [0088], The machine learning programs or algorithms may also include natural language processing, semantic analysis, automatic reasoning, regression analysis, support vector machine (SVM) analysis, decision tree analysis, random forest analysis, K-Nearest neighbor analysis, naïve Bayes analysis, clustering, reinforcement learning, and/or other machine learning algorithms and/or techniques).
Regarding to claim 10, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the large language model module comprises a drift asset allocation to optimize of the asset allocation in an enterprise (para [0068], the machine learning training model 134 may fine-tune an existing GPT model. The existing model may be, for example, an open source large language model (LLM) (e.g., Large Language Model Meta AI (LLaMA)) or another small-parameter or large-parameter LLM. The fine-tuning may include grounding a pre-trained model by providing additional training examples using a zero-shot or few-shot learning strategy).
Regarding to claim 11, Pyle discloses the non-transitory computer-readable medium of claim 1, wherein the recommendation is materialized based on input data sensed by the processor on whether the recommendation is accepted by a user enterprise premises data source (para [0071], The customers may also request and receive information related to the rule group 502-A. For example, the customer may request information related to security recommendations, as depicted in FIG. 5I. The GPT model may respond by displaying one or more recommendations that the user may perform in order to improve the user's cloud maturity score. The recommendations may be ranked and prioritized as discussed herein).
Claims 12-15 are written in method and contain the same limitations found in claims 1, 5-6, and 8 described above, therefore are rejected by the same rationale.
Regarding to claims 16-20, Pyle discloses a system to generate decision recommendations to optimize asset allocation in an enterprise, the system comprising:
at least one hardware processor; and software that is configured to, when executed by the at least one hardware processor (para [0059], The client computing device 102 may include a processor and a network interface controller (NIC). The processor may include any suitable number of processors and/or processor types, such as CPUs and one or more graphics processing units (GPUs). The processor is configured to execute software instructions stored in a memory.) and perform the method described in claims 1, 5-6, and 8-9 described above, therefore are rejected by the same rationale.
Response to Arguments/Amendment
5. Applicant's arguments with respect to claims 1-20 have been considered but are moot because the new ground of rejection.
I. Claim Rejections - 35 USC § 101
Claims 1-20 are eligible because the claims recite significantly more than the abstract idea and integrate the abstract idea into a practical application, specially the limitations “wherein the large language model module is enhanced using retrieval- augmented generation (RAG) that combines generative capabilities with a retrieval mechanism to retrieve relevant information from a dataset and generate contextually relevant responses through text processing, thereby improving the large language model module's ability to provide insight and recommendation for asset allocation; and wherein the large language model module processes text input to perform intent detection and entity recognition, and creates the intent into a schema mapping to retrieve and detect assets within the enterprise, and constructs a query to retrieve requested data based on the schema mapping”, provide improvements to the large language model (LLM) by combining with the Retrieval-Augmented Generation (RAG). This RAG enhancement is a specific technical architecture-not merely a generic invocation of AI technology-that improves the LLM module's performance by combining generative capabilities with a retrieval mechanism. As described in the specification, "RAG can enhance LLM module's 420 performance by combining generative capabilities with a retrieval mechanism" and the "enhanced LLM module 420 retrieves relevant information from dataset 312." Specification, at [90]. The retrieved information is then used to "generate a more accurate and contextually relevant response through text processing." Specification, at [91]. By integrating external dataset sources, RAG improves the LLM module's ability to provide insight and recommendation for asset allocation and contributes to the continuous improvement of the LLM module. Therefore, the claims are eligible.
According, the 101 rejection has been withdrawn.
II. Claim Rejections - 35 USC § 102
Applicant’s arguments with respect to claims 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. New ground of 103 rejection described above.
Conclusion
6. 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 extension fee 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 date of this final action.
7. Claims 1-20 are rejected.
8. The prior arts made of record and not relied upon are considered pertinent to applicant's disclosure:
Khosla et al. (US 2025/0005057) disclose the LLM may be a trained machine learning model utilizing Retrieval Augmented Generation (RAG) techniques to generate answers using semantics (e.g., in addition to or alternatively to lexical techniques) to answer the question (see para [0013]).
Kislal et al. (US 2024/0012842) disclose RAG language model—the Retrieval-Augmented Generation (RAG) language model combines pre-trained parametric and non-parametric memory for language generation (see para [0058]).
9. Any inquiry concerning this communication or earlier communications from the examiner should be directed to examiner NGA B NGUYEN whose telephone number is (571) 272-6796. The examiner can normally be reached on Monday-Friday 7AM-5PM.
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, Beth Boswell can be reached on (571) 272-6737. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NGA B NGUYEN/Primary Examiner, Art Unit 3625 May 29, 2026
Prosecution Insights