Analyzing enterprise activity to identify patentable subject matter and create a corresponding patent disclosure

A computing device comprising:

• one or more processors;
• and
• one or more non-transitory computer readable media storing instructions executable by the one or more processors to perform operations comprising:
• gathering, from a private network, data in a private network that is associated with an enterprise, the data comprising machine-generated digital records comprising at least one of runtime logs, system telemetry, or execution traces automatically produced during software development and runtime operations within the enterprise;
• determining a content type of individual data items in the data;
• selecting a content-specific encoder from a plurality of encoders based on the content type;
• applying the content-specific encoder to transform the data item into a corresponding vector embedding;
• determining, by an artificial intelligence and based on one or more policies, a probability that individual vector embeddings in the set of vector embeddings include a patentable idea;
• adding the patentable idea to a set of patentable ideas based on determining that the probability is greater than a predetermined patentability threshold;
• performing by a reasoning and acting agent, a search of external databases for individual patentable ideas in the set of patentable ideas to create search results, wherein the reasoning and acting agent executes iterative thought-action-observation cycles in which individual cycles comprise:
• generating a search query based on a current reasoning state,
• submitting the search query to the external databases,
• evaluating results received from the external databases, and
• determining whether to refine the query for a subsequent cycle based on evaluating the results;
• determining a score for individual patentable ideas in the set of patentable ideas based on the search results, the score comprising a weighted combination of a novelty score, an enablement score, and a clarity score representing respective section-level quality metrics of a candidate patent disclosure;
• creating, by a large language model, a patent disclosure document for individual patentable ideas based at least in part on determining that the score is greater than a predetermined patent disclosure threshold;
• providing the patent disclosure document to a patent review committee;
• receiving comments from individual members of the patent review committee;
• updating the score to create an updated score by applying an iterative reinforcement learning update rule that incorporates a reward signal derived from the comments, wherein the reward signal comprises a weighted combination of at least a precision metric, a recall metric, and a false positive rate with respect to patentable idea detection;
• and
• adjusting, by a reinforcement learning controller and based on the updated score, at least one operating parameter of the artificial intelligence comprising the predetermined patentability threshold, to improve detection of patentable ideas across subsequent data-gathering cycles.

One or more non-transitory computer readable media storing instructions executable by one or more processors to perform operations comprising:

• gathering, from a private network, data in a private network that is associated with an enterprise, the data comprising machine-generated digital records comprising at least one of runtime logs, system telemetry, or execution traces automatically produced during software development and runtime operations within the enterprise;
• determining a content type of individual data items in the data;
• selecting a content-specific encoder from a plurality of encoders based on the content type;
• applying the content-specific encoder to transform the data item into a corresponding vector embedding;
• determining, by an artificial intelligence and based on one or more policies, a probability that individual vector embeddings in the set of vector embeddings include a patentable idea;
• adding the patentable idea to a set of patentable ideas based on determining that the probability is greater than a predetermined patentability threshold;
• performing by a reasoning and acting agent, a search of external databases for individual patentable ideas in the set of patentable ideas to create search results, wherein the reasoning and acting agent executes iterative thought-action-observation cycles in which individual cycles comprise:
• generating a search query based on a current reasoning state,
• submitting the search query to the external databases,
• evaluating results received from the external databases, and
• determining whether to refine the query for a subsequent cycle based on evaluating the results;
• determining a score for individual patentable ideas in the set of patentable ideas based on the search results, the score comprising a weighted combination of a novelty score, an enablement score, and a clarity score representing respective section-level quality metrics of a candidate patent disclosure;
• creating, by a large language model, a patent disclosure document for individual patentable ideas based at least in part on determining that the score is greater than a predetermined patent disclosure threshold;
• providing the patent disclosure document to a patent review committee;
• receiving comments from individual members of the patent review committee;
• updating the score to create an updated score by applying an iterative reinforcement learning update rule that incorporates a reward signal derived from the comments, wherein the reward signal comprises a weighted combination of at least a precision metric, a recall metric, and a false positive rate with respect to patentable idea detection;
• and
• adjusting, by a reinforcement learning controller and based on the updated score, at least one operating parameter of the artificial intelligence comprising the predetermined patentability threshold, to improve detection of patentable ideas across subsequent data-gathering cycles.