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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 05/18/2026 has been entered.
Response to Arguments
Applicant's arguments filed 05/18/2026 have been fully considered but they are not persuasive.
Regarding the 103 rejections, applicant's arguments filed with respect to the prior art rejections have been fully considered but they are moot. Applicant has amended the claims to recite new combinations of limitations. Applicant's arguments are directed at the amendment. Please see below for new grounds of rejection, necessitated by Amendment.
Claim Rejections - 35 USC § 112: Indefiniteness
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.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 1, the claim recites the limitation wherein: the prompt comprises a structured representation of the at least a response data element that is configured to constrain generation of the time-correlated data output. There is insufficient antecedent basis for this limitation in the claim because the term “the time-correlated data output” lacks antecedent basis. For the purposes of examination, a time-correlated data output is an output from a model that has an associated time element.
Regarding claims 2-10, the claims are rejected for at least their dependence to claim 1.
Regarding claims 11-20, the claims are similar to claims 1-10 and are rejected under the same rationales.
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 1-5, 9, 11-15, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Lombard, et al., US Pre-Grant Publication US20240362718A1 (“Lombard”) in view of De Zarza, et al., Non-Patent Literature “Optimized Financial Planning: Integrating Individual and Cooperative Budgeting Models with LLM Recommendations” (“De Zarza”) and further in view of Arriaga, et al., US Pre-Grant Publication US20250053674A1 (“Arriaga”) and Gaur, et al., US Pre-grant Publication US20250391547A1 (“Gaur”).
Regarding claim 1 and analogous claim 11, Lombard discloses:
An apparatus for generating time-correlated data outputs, the apparatus comprising: a processor; and a memory communicatively connected to the processor, wherein the memory contains instructions configuring the processor to: (Lombard, ⁋4, “In an aspect, a platform-agnostic digital assistant apparatus is disclosed [An apparatus for generating time-correlated data outputs,]. The apparatus includes a processor [the apparatus comprising: at a processor;] and a memory communicatively connected to the processor, wherein the memory contains instructions configuring the processor to [a memory communicatively connected to the processor, wherein the memory contains instructions configuring the processor to:]”).
receive from an entity an input…, the input…comprising (Lombard, ⁋26, “The processor 108 may be further configured to utilize the communication channel 128 to exchange first information 140 related to a user between the processor 108 and the entity [receive from an entity] 132. “First information,” [an input…, the input…comprising] as used herein, is defined as information exchanged between processor 108 and entity 132 or information transmitted by entity 132.”).
at least a query attribute and at least a resource attribute; (Lombard, ⁋27, “processor 108 is further configured to extract at least one user datum from the first information 140. “User datum,” as used herein, is defined as an element of information related to a user. User datum 144 may include user identifying information [at least a query attribute] (such as name, address, social security number, age, height, weight, health, and the like)…insurance information relating to a user [and at least a resource attribute;] (such as current policy coverage, policy dates, policy costs, a deductible, a policy rate, claims filed, and the like)”).
receive response generation training data, wherein the response generation training data comprises exemplary input…correlated with exemplary response data elements; (Lombard, ⁋51, “With continued reference to FIG. 1, once language classification training data [receive response generation training data,] is created or received by processor 108, a dynamic response machine learning model may be trained using the language classification training data; classification data is interpreted as having inputs with labeled outputs (i.e. wherein the response generation training data comprises exemplary input…correlated with exemplary response data elements;) by processor 108, machine learning module 116, language processing module 124, or another device. In a non-limiting embodiment, language classification training data is submitted to a machine-learning model, which generates a trained dynamic response machine learning model based on the correlated relationship or relationships between language elements.”).
determine at least a response data element using the trained response generation machine-learning model, as a function of the input…, (Lombard, ⁋52, “processor 108 may be configured to utilize a trained dynamic response machine learning model [using a response generation machine-learning model trained on response generation training data,] to generate dynamic response [determine at least a response data element] information based on portions of first information [as a function of the input…,] 140 generated by entity 132. For example, a trained dynamic response machine learning model may receive a sentence or similar written communication from entity 132 as an input. The trained dynamic response machine learning model may then generate written output corresponding to the information and style of the input from entity 132,”).
While Lombard teaches the use of time-correlated output generation using machine learning, Lombard does not explicitly teach:
…input data structures…
sanitize the response generation training data, wherein sanitizing the response generation training data comprises: determining by the processor that at least one training data entry of the response generation training data has a signal to noise ratio below a threshold value; and removing the at least one training data entry from the response generation training data to create sanitized response generation training data;
train a response generation machine-learning model as a function of the sanitized response generation training data;
… wherein determining the at least a response data element comprises: computing at least a response metric as a function of the at least a query attribute and the at least a resource attribute; and determining the at least a response data element as a function of the at least a response metric;
generate, using a large language model (LLM), a prompt as a function of the at least a response data element which corresponds to the at least a response metric comprising a financial responsibility outline for a user, wherein: the prompt comprises a structured representation of the at least a response data element that is configured to constrain generation of the time-correlated data output;
and the LLM comprises an attention mechanism configured to dynamically quantify features of the input data, wherein the attention mechanism is configured to: locate relevant information within a sentence; and predict a next word based on context vectors;
create at least a time-correlated data output as a function of the prompt, wherein creating the at least a time-correlated data output comprises generating a plurality of data elements distributed across a plurality of time points according to at least one temporal attribute derived from the prompt;
and display, using a graphical user interface, the at least a time-correlated data output, wherein the graphical user interface comprises at least an event handler graphic.
De Zarza teaches:
…input data structure… (De Zarza, pg. 107, “response = openai.ChatCompletion.create( model="gpt-4-0613", messages=[ {"role": "system", "content": "You are an expert financial planner with expertise in life-cycle models and cooperative budgeting for households. Provide a comprehensive recommendation based on the combined financial data that considers consumption smoothing over the entire life-cycle."}, {"role": "user", "content": f"Here’s our combined household financial data, along with some future projections and life events: {household_data_string}. Can you recommend how we should collaboratively budget our expenses?"}]); this prompt to a LLM contains a user’s financial data and query within an array (i.e. …input data structure…)”).
…wherein determining the at least a response data element comprises: computing at least a response metric as a function of the at least a query attribute and the at least a resource attribute; and determining the at least a response data element as a function of the at least a response metric; (De Zarga, pg. 100, “To empower users in optimizing their interactions with the LLM, we can also introduce a guided prompt optimization feature within our LLM-informed budgeting system. This enhancement can educate users on constructing effective prompts, thereby improving the quality of the financial recommendations that they receive; an LLM-informed budgeting system is interpreted as having a required goal that needs to be satisfied, or a response metric (i.e. wherein determining the at least a response data element comprises: computing at least a response metric…and determining the at least a response data element as a function of the at least a response metric;). Here is how we can incorporate this guidance into the user experience: Prompt Templates: We provide users with a collection of prompt templates that are designed based on successful financial inquiry patterns. These templates serve as starting points for users to articulate their financial data and questions [as a function of the at least a query attribute and the at least a resource attribute;].”).
generate, using a large language model (LLM), a prompt as a function of the at least a response data element which corresponds to the at least a response metric comprising a financial responsibility outline for a user, wherein: the prompt comprises a structured representation of the at least a response data element that is configured to constrain generation of the time-correlated data output; (De Zarza, pg. 107, “response = openai.ChatCompletion.create( model="gpt-4-0613" [generate, using a large language model (LLM),], messages=[ {"role": "system", "content": "You are an expert financial planner with expertise in life-cycle models and cooperative budgeting for households. Provide a comprehensive recommendation based on the combined financial data that considers consumption smoothing over the entire life-cycle [a prompt as a function of the at least a response data element which corresponds to the at least a response metric comprising a financial responsibility outline for a user, wherein: the prompt comprises a structured representation of the at least a response data element that is configured to constrain generation of the time-correlated data output;]."}, {"role": "user", "content": f"Here’s our combined household financial data, along with some future projections and life events: {household_data_string}. Can you recommend how we should collaboratively budget our expenses?"}])”).
create at least a time-correlated data output as a function of the prompt, wherein creating the at least a time-correlated data output comprises generating a plurality of data elements distributed across a plurality of time points according to at least one temporal attribute derived from the prompt; (De Zarza, pg. 108, “Using the life-cycle model involves setting your saving and spending habits with a long-term vision. Consumption smoothing is a key part of this, aiming to allocate resources in a way that maintains a consistent standard of living throughout your life. Here are my recommendations based on both your income and projected life events [create at least a time-correlated data output as a function of the prompt,]: 1. **Start an Emergency Fund**: Save a portion of your combined income consistently until you have about three to six months’ worth of living expenses. Considering your incomes and basic monthly expenses, which total $2150, I’d suggest allocating at least $500/month for this initially. 2. **Create a Child Fund**: Since you’re planning to have a child in 3 years, creating a dedicated child fund should be a priority. It will help cover everything from pregnancy-related medical costs to future schooling expenses. Let’s allocate $300/month for this fund [wherein creating the at least a time-correlated data output comprises generating a plurality of data elements distributed across a plurality of time points according to at least one temporal attribute derived from the prompt;].”).
and display, using a graphical user interface, the at least a time-correlated data output, wherein the graphical user interface comprises at least an event handler graphic. (De Zarza, pg. 110 and see the LLM output on pg. 108, “In the process of developing our LLM financial recommender system, we mapped the interactions between various components to ensure a seamless user experience and robust performance. Figure 3 illustrates the sequence of events from the moment a user inputs their financial data to the receipt of personalized recommendations and subsequent budget optimization. This diagram delineates the system’s workflow, highlighting the role of the LLM in analyzing financial data and generating tailored advice [and display, using a graphical user interface, the at least a time-correlated data output, wherein the graphical user interface comprises at least an event handler graphic.].”).
Lombard and De Zarza are both in the same field of endeavor (i.e. AI digital assistants). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard and De Zarza to teach the above limitation(s). The motivation for doing so is that using a LLM improves model outputs by leveraging the LLM’s ability to personalize and tailor responses per user (cf. De Zarza, pg. 97, “In the modern era of data-driven solutions, using an LLM to provide financial advice offers an innovative approach to household budgeting. Given the intricate nature of individual spending habits and the myriad of ways in which funds can be allocated, using the power of AI can significantly enhance personalized financial recommendations.”).
While Lombard in view of De Zarza teaches time-correlated output generation using an LLM, prompting, and data structures, the combination does not explicitly teach:
sanitize the…training data, wherein sanitizing the…training data comprises: determining by the processor that at least one training data entry of the…training data has a signal to noise ratio below a threshold value; and removing the at least one training data entry from the…training data to create sanitized…training data;
train a…machine-learning model as a function of the sanitized…training data;
and the LLM comprises an attention mechanism configured to dynamically quantify features of the input data, wherein the attention mechanism is configured to: locate relevant information within a sentence; and predict a next word based on context vectors;
Arraiga teaches:
sanitize the…training data, wherein sanitizing the…training data comprises: determining by the processor that at least one training data entry of the…training data has a signal to noise ratio below a threshold value; and removing the at least one training data entry from the…training data to create sanitized…training data; (Arriaga, ⁋117, “a value that is more than a threshold number of standard deviations away from an average, mean, or expected value, for instance, may be eliminated [removing the at least one training data entry from the…training data to create sanitized…training data;]. Alternatively or additionally, one or more training examples may identify as having poor quality data, where “poor quality” is defined as having a signal to noise ratio below a threshold value [sanitize the…training data, wherein sanitizing the…training data comprises: determining by the processor that at least one training data entry of the…training data has a signal to noise ratio below a threshold value;].” and Arriaga, ⁋117, “Still referring to FIG. 4, computer, processor, and/or module may be configured to sanitize training data [determining by the processor].”).
train a…machine-learning model as a function of the sanitized…training data; (Arriaga, ⁋117, ““Sanitizing” training data, as used in this disclosure, is a process whereby training examples are removed that interfere with convergence of a machine-learning model and/or process to a useful result [train a…machine-learning model as a function of the sanitized…training data;].”).
Lombard, in view of De Zarza, and Arriaga are both in the same field of endeavor (i.e. machine learning). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard, in view of De Zarza, and Arriaga to teach the above limitation(s). The motivation for doing so is that sanitizing training data can remove the negative impacts of poor quality training samples (cf. Arriaga, ⁋117, ““Sanitizing” training data, as used in this disclosure, is a process whereby training examples are removed that interfere with convergence of a machine-learning model and/or process to a useful result.”).
While Lombard in view of De Zarza and Arriaga teaches time-correlated output generation using LLMs, input data structures, and data sanitization, the combination does not explicitly teach:
and the LLM comprises an attention mechanism configured to dynamically quantify features of the input data, wherein the attention mechanism is configured to: locate relevant information within a sentence; and predict a next word based on context vectors;
Gaur teaches and the LLM comprises an attention mechanism configured to dynamically quantify features of the input data, wherein the attention mechanism is configured to: locate relevant information within a sentence; and predict a next word based on context vectors; (Gaur, ⁋36, “The attention mechanisms help neural networks to learn the context of words in the sequences of words [and the LLM comprises an attention mechanism configured to dynamically quantify features of the input data,]. An attention mechanism operates by breaking down a set of input data, such as a sentence or sequence of words or tokens, into keys, queries, and values. Keys represent elements of the input data that provide information about what to pay attention to [wherein the attention mechanism is configured to: locate relevant information within a sentence; and].” and Gaur, ⁋37, “In operation, the LLM receives a natural language prompt as input data and generates a sequence of words in natural language by predicting a next word, or sequence of words, based on the textual and grammatical patterns learned by the LLM during training [predict a next word based on context vectors;].”).
Lombard, in view of De Zarza and Arriaga, and Gaur are both in the same field of endeavor (i.e. machine learning). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard, in view of De Zarza and Arriaga, and Gaur to teach the above limitation(s). The motivation for doing so is that an attention mechanism for sentence understanding improves the contextual information gained by a model (cf. Gaur, ⁋36, “The attention mechanisms help neural networks to learn the context of words in the sequences of words. An attention mechanism operates by breaking down a set of input data, such as a sentence or sequence of words or tokens, into keys, queries, and values. Keys represent elements of the input data that provide information about what to pay attention to.”).
Regarding claim 2 and analogous claim 12, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. De Zarza further teaches wherein determining the at least, a response data element comprises: computing the at least a response metric as a function of the at least a query attribute and the at least a resource attribute; and determining the at least a response data element as a function of the at least a response metric wherein the at least a response metric is embedded within the at least a response data element and propagated to the prompt to constrain generation of the time-correlated data output. (De Zarga, pg. 100, “To empower users in optimizing their interactions with the LLM, we can also introduce a guided prompt optimization feature within our LLM-informed budgeting system. This enhancement can educate users on constructing effective prompts, thereby improving the quality of the financial recommendations that they receive; an LLM-informed budgeting system is interpreted as having a required goal that needs to be satisfied, or a response metric (i.e. wherein determining the at least, a response data element comprises: computing the at least a response metric…and determining the at least a response data element as a function of the at least a response metric). Here is how we can incorporate this guidance into the user experience: Prompt Templates: We provide users with a collection of prompt templates that are designed based on successful financial inquiry patterns. These templates serve as starting points for users to articulate their financial data and questions [as a function of the at least a query attribute and the at least a resource attribute;].”, and De Zarza, pg. 107, “response = openai.ChatCompletion.create( model="gpt-4-0613", messages=[ {"role": "system", "content": "You are an expert financial planner with expertise in life-cycle models and cooperative budgeting for households. Provide a comprehensive recommendation based on the combined financial data that considers consumption smoothing over the entire life-cycle [wherein the at least a response metric is embedded within the at least a response data element and propagated to the prompt to constrain generation of the time-correlated data output.]."}, {"role": "user", "content": f"Here’s our combined household financial data, along with some future projections and life events: {household_data_string}. Can you recommend how we should collaboratively budget our expenses?"}])”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of De Zarza with the teachings of Lombard, Arriaga, and Gaur for the same reasons disclosed in claim 1.
Regarding claim 3 and analogous claim 13, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. De Zarza teaches creating the at least a time-correlated data output as seen in claim 1. Lombard further teaches:
receiving supplemental input data; (Lombard, ⁋65, “processor 108 may be further configured to determine an indemnity outlay [receiving supplemental input data;] based on the user datum 144. As used herein, “indemnity outlay” is defined as a currency value related to insurance.”).
identifying at least a temporal attribute from the supplemental input data; and creating the at least a time-correlated data output as a function of the at least a temporal attribute. (Lombard, ⁋65, “For example, processor 108 may determine that first information 140 includes user datum 144 indicating that a user's vehicle was totaled in an accident and will be written off as a total loss. Processor 108 may use this information to retrieve data related to a user's insured vehicle including value of the vehicle. Processor 108 may then determine an indemnity outlay to be paid out to a user; being paid out to the user is interpreted as creating a time-correlated output (i.e. and creating the at least a time-correlated data output as a function of the at least a temporal attribute.) corresponding to the value of the insured vehicle. An indemnity outlay may be a currency value corresponding to a claim amount, a deductible, a monthly insurance rate [identifying at least a temporal attribute from the supplemental input data;], a lawsuit settlement, and the like.”).
Regarding claim 4 and analogous claim 14, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. Arraiga teaches the sanitized training dataset as seen in claim 1. Lombard further teaches:
wherein training the response generation machine-learning model comprises: iteratively training the response generation machine-learning model as a function of the …response generation training data; and (Lombard, ⁋51, “Connections between nodes may be created via the process of “training” the network, in which elements from language classification training data [as a function of the…response generation training data;] set are applied to the input nodes, a suitable training algorithm (such as Levenberg-Marquardt, conjugate gradient, simulated annealing, or other algorithms) is then used to adjust the connections and weights between nodes in adjacent layers of the neural network to produce the desired natural language output values at the output nodes. This process is sometimes referred to as deep learning; deep learning is interpreted as an iterative training process [i.e. iteratively training the response generation machine-learning model].”).
synthesizing the at least a response data element using the trained response generation machine-learning model. (Lombard, ⁋52, “processor 108 may be configured to utilize a trained dynamic response machine learning model to generate dynamic response information [synthesizing the at least a response data element using the trained response generation machine-learning model.] based on portions of first information 140 generated by entity 132.”).
Regarding claim 5 and analogous claim 15, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. Lombard further teaches wherein receiving the input data structure comprises: querying a data repository using one or more elements of the input data structure; and validating the entity associated with the input data structure as a function of an outcome of the query. (Lombard, ⁋29, “Processor 108 may then analyze the replied portion of first information [using one or more elements of the input data structure;] 140 generated by entity 132 by parsing the reply and searching for a number consisting of a predetermined number of digits, such as eight. The processor 108 may then compare the account number to a database [querying a data repository] associating account numbers and user identifying information and determine that account number 55375914 belongs to user Bill Jones and is therefore a valid user datum 144 [and validating the entity associated with the input data structure as a function of an outcome of the query.].”).
Regarding claim 9 and analogous claim 19, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. De Zarza further teaches wherein displaying the at least a time-correlated data output comprises displaying the at least an event handler graphic as a function of the at least a time-correlated data output. (De Zarza, pg. 110 and see the LLM output on pg. 108, “In the process of developing our LLM financial recommender system, we mapped the interactions between various components to ensure a seamless user experience and robust performance. Figure 3 illustrates the sequence of events from the moment a user inputs their financial data to the receipt of personalized recommendations and subsequent budget optimization. This diagram delineates the system’s workflow, highlighting the role of the LLM in analyzing financial data and generating tailored advice; the simulated output on pg. 108 shows that the recommended budget is displayed to the user in a chat window (i.e. wherein displaying the at least a time-correlated data output comprises displaying the at least an event handler graphic as a function of the at least a time-correlated data output.).”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of De Zarza with the teachings of Lombard, Arriaga, and Gaur for the same reasons disclosed in claim 1.
Claims 6-7 and 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Lombard, et al., US Pre-Grant Publication US20240362718A1 (“Lombard”) in view of De Zarza, et al., Non-Patent Literature “Optimized Financial Planning: Integrating Individual and Cooperative Budgeting Models with LLM Recommendations” (“De Zarza”) and further in view of Arriaga, et al., US Pre-Grant Publication US20250053674A1 (“Arriaga”), Gaur, et al., US Pre-grant Publication US20250391547A1 (“Gaur”), and Zahora, et al., US Pre-Grant Publication US20220309592A1 (“Zahora”).
Regarding claim 6 and analogous claim 16, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. The combination also teaches wherein creating the at least a time-correlated data output as seen in claim 1. While the combination teaches the use of time-correlated output generation using machine learning, the combination does not explicitly teach assigning a compatibility metric to each time-correlated data output of a plurality of time-correlated data outputs; and ranking the plurality of time-correlated data outputs as a function of the compatibility metric.
Zahora teaches assigning a compatibility metric to each time-correlated data output of a plurality of time-correlated data outputs; and ranking the plurality of time-correlated data outputs as a function of the compatibility metric. (Zahora, ⁋127, “In some implementations, the predictive analytics platform 102 includes a payment pattern application engine 226 configured to apply payer payment patterns produced by the payer payment pattern engine 222 and/or patient payer patterns produced by the patient payment pattern engine 224 to calculate payment estimations. Further, in some embodiments, the payment pattern application engine 226 determines a confidence level or rating associated with the payment estimate [assigning a compatibility metric to each time-correlated data output of a plurality of time- correlated data outputs;]. The confidence level or rating, in some examples, can include a percentage confidence, a ranked confidence (e.g., on a scale of 1 to X), and/or a ratings confidence (e.g., low, medium, high) [ranking the plurality of time-correlated data outputs as a function of the compatibility metric.].”).
Lombard, in view of De Zarza, Arriaga, and Gaur, and Zahora are both in the same field of endeavor (i.e. machine learning data analysis). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard, in view of De Zarza, Arriaga, and Gaur, and Zahora to teach the above limitation(s). The motivation for doing so is that ranking results streamlines the processing of claims (cf. Zahora, ⁋3, “The billing company or medical provider administration platform, for example, may streamline the complexities of claims processing and records keeping on behalf of the medical provider.”).
Regarding claim 7 and analogous claim 17, Lombard in view of De Zarza, Arriaga, Gaur, and Zahora teaches the apparatus of claim 6. Zahora further teaches:
wherein the processor is further configured to: compare one or more elements of the input data structure against one or more prerequisite metrics; and (Zahora ⁋117, “The predictive analytics platform 102 and/or a revenue maximizer portal 110 application, for example, may perform calculations to determine payment estimates based upon the values and likelihoods contained within the payer payment pattern(s) ; values and likelihoods within payment patterns are interpreted as prerequisite metrics (i.e compare one or more elements of the input data structure against one or more prerequisite metrics;) and patient payer pattern (e.g., the payment patterns & payment estimates 132 of FIG. 1A). The calculations may include statistically combining costs and likelihoods to produce a final estimate.”).
modify the ranked plurality of time-correlated data outputs as a function of comparing the one or more elements of the input data structure against the one or more prerequisite metrics. (Zahora, ⁋127, “In some implementations, the predictive analytics platform 102 includes a payment pattern application engine 226 configured to apply payer payment patterns produced by the payer payment pattern engine 222 and/or patient payer patterns produced by the patient payment pattern engine 224 to calculate payment estimations. Further, in some embodiments, the payment pattern application engine 226 determines a confidence level or rating associated with the payment estimate; since the values and likelihoods of the payment patterns is interpreted as the prerequisite metrics, the payment pattern engine performing the ranking is modifying the rankings based on the values and likelihoods (i.e. modify the ranked plurality of time-correlated data outputs as a function of comparing the one or more elements of the input data structure against the one or more prerequisite metrics.). The confidence level or rating, in some examples, can include a percentage confidence, a ranked confidence (e.g., on a scale of 1 to X), and/or a ratings confidence (e.g., low, medium, high).”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Zahora with the teachings of Lombard, De Zarza, Arraiga, and Gaur for the same reasons disclosed in claim 6.
Claims 8 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Lombard, et al., US Pre-Grant Publication US20240362718A1 (“Lombard”) in view of De Zarza, et al., Non-Patent Literature “Optimized Financial Planning: Integrating Individual and Cooperative Budgeting Models with LLM Recommendations” (“De Zarza”) and further in view of Arriaga, et al., US Pre-Grant Publication US20250053674A1 (“Arriaga”), Gaur, et al., US Pre-grant Publication US20250391547A1 (“Gaur”), and Wu, et al., Non-Patent Literature “A survey on large language models for recommendation” (“Wu”).
Regarding claim 8 and analogous claim 18, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. Gaur further teaches synthesizing the prompt using the LLM trained on a plurality of training examples, (Gaur, ⁋35, “LLMs contain hundreds of billions of parameters trained on multiple terabytes of text. LLMs are trained to receive natural language as an input. LLMs typically generate natural language as an output [synthesizing the prompt using the LLM trained on a plurality of training examples,].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Gaur with the teachings of Lombard, De Zarza, and Arraiga for the same reasons disclosed in claim 1.
While, the combination teaches the use of time-correlated output generation using an LLM, the combination does not explicitly teach wherein training the LLM comprises: pretraining the LLM on a general set of training examples; and fine-tuning the LLM on a special set of training examples, wherein the general and the special set of training examples are subsets of the plurality of training examples.
Wu teaches wherein training the LLM comprises: pretraining the LLM on a general set of training examples; and fine-tuning the LLM on a special set of training examples, wherein the general and the special set of training examples are subsets of the plurality of training examples. (Wu, pg. 60, “The process of fine-tuning involves initializing the pre-trained language model with its learned parameters [wherein training the LLM comprises: pretraining the LLM on a general set of training examples;] and then training it on a recommendation-specific dataset. This dataset typically includes user-item interactions, textual descriptions of items, user profiles, and other relevant contextual information. During fine-tuning, the model’s parameters are updated based on the task-specific data [fine-tuning the LLM on a special set of training examples,], allowing it to adapt and specialize for target recommendation tasks. The learning objectives in the pre-training and fine-tuning stages can vary, as they are aimed at different optimization targets; all training datasets are interpreted as being part of the plurality of training examples thus the pre-training dataset and the fine-tuning dataset are interpreted as being part of the plurality of training examples (i.e. wherein the general and the special set of training examples are subsets of the plurality of training examples.).”).
Lombard, in view of De Zarza, Arriaga, and Gaur, and Wu are both in the same field of endeavor (i.e. response/recommendation generation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard, in view of De Zarza, Arriaga, and Gaur, and Wu to teach the above limitation(s). The motivation for doing so is that using large language models improve recommendation tasks (cf. Wu, pg. 2, “The key advantage of incorporating LLMs into recommendation systems lies in their ability to extract high-quality representations of textual features and leverage the extensive external knowledge encoded within them [1].”).
Claims 10 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Lombard, et al., US Pre-Grant Publication US20240362718A1 (“Lombard”) in view of De Zarza, et al., Non-Patent Literature “Optimized Financial Planning: Integrating Individual and Cooperative Budgeting Models with LLM Recommendations” (“De Zarza”) and further in view of Arriaga, et al., US Pre-Grant Publication US20250053674A1 (“Arriaga”), Gaur, et al., US Pre-grant Publication US20250391547A1 (“Gaur”), and Anthony, et al., US Pre-Grant Publication US20250005224A1 (“Anthony”).
Regarding claim 10 and analogous claim 20, Lombard in view of De Zarza, Arriaga, and Gaur teaches the apparatus of claim 1. The combination teaches wherein displaying the at least a time-correlated data output as seen in claim 1. While the combination teaches the use of time-correlated output generation using machine learning, the combination does not explicitly teach generating a color-coded visualization as a function of the at least a time-correlated data output; and displaying the generated color-coded visualization.
Anthony teaches generating a color-coded visualization as a function of the at least a time-correlated data output; and displaying the generated color-coded visualization. (Anthony, ⁋52, “Component 213 receives response 230 (or input 229) from module 207 and generates user interface 235 based on response 230 (or input 229). If response 230 includes indication 233, then user interface 235 may present [displaying the generated color-coded visualization.] the content of indication 233 as a confidence level having a rating (e.g., 65% confident of the response's accuracy), having a color-coded score (e.g., a green color indicates high confidence, a red color indicates a low confidence, etc.) [generating a color-coded visualization as a function of the at least a time-correlated data output;], and the like.”).
Lombard, in view of De Zarza, Arriaga, and Gaur, and Anthony are both in the same field of endeavor (i.e. response/recommendation generation). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Lombard, in view of De Zarza, Arriaga, and Gaur, and Anthony to teach the above limitation(s). The motivation for doing so is color coordination aids in conveying information to the user (cf. Anthony, ⁋52, “If response 230 includes indication 233, then user interface 235 may present the content of indication 233 as a confidence level having a rating (e.g., 65% confident of the response's accuracy), having a color-coded score”).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICHOLAS S WU whose telephone number is (571)270-0939. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 pm 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, Michelle Bechtold can be reached on 571-431-0762. 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.
/N.S.W./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights