Accelerated knowledge discovery for knowledge base

A system, comprising:

• at least one computer processor;
• a large language model (LLM), executing on the at least one computer processor;
• a document improvement engine, executing on the at least one computer processor, and configured to:
• iteratively executing a pre-processing document processing loop until a second document readiness score is greater than a document readiness threshold, comprising:
• analyzing, by the LLM, a raw document based on a raw document type of the raw document, to obtain a document readiness score of the raw document,
• responsive to the document readiness score being less than the document readiness threshold, performing operations comprising:
• modifying, by a LLM, the raw document in accordance with a modification recommendation generating based on evaluating the raw document against a document rubric of the raw document type, to obtain a modified document,
• augmenting, by the LLM, the modified document with a semantic context to obtain a discoverable document, and
• generating by the LLM, the second document readiness score corresponding to the discoverable document;
• and
• a document discovery engine, comprising an embedding generator, a searcher, and an answer generator, and executing on the at least one computer processor, wherein the document discovery engine is configured to execute a post-processing document processing phase, comprising:
• obtaining a discoverable document embedding corresponding to the discoverable document that is semantically similar to a user query embedding of a user query,
• generating an answer, by the LLM, to the user query from the discoverable document embedding, and
• triggering the execution of the pre-processing document processing loop to perform updating the discoverable document with the semantic context.

A non-transitory computer readable medium comprising instructions for causing a computing system to perform operations comprising:

• iteratively executing a pre-processing document processing loop until a second document readiness score is greater than a document readiness threshold, comprising:
• analyzing, by a large language model (LLM), a raw document based on a raw document type of the raw document, to obtain a first document readiness score of the raw document,
• responsive to the first document readiness score being less than the document readiness threshold, performing operations comprising:
• modifying, by a LLM, the raw document in accordance with a modification recommendation generating based on evaluating the raw document against a document rubric of the raw document type, to obtain a modified document,
• augmenting, by the LLM, the modified document with a semantic context to obtain a discoverable document, and
• generating by the LLM, the second document readiness score corresponding to the discoverable document;
• and
• executing a post-processing document processing phase, comprising:
• obtaining a discoverable document embedding corresponding to the discoverable document that is semantically similar to a user query embedding of a user query,
• generating an answer, by the LLM, to the user query from the discoverable document embedding, and
• triggering the execution of the pre-processing document processing loop to perform updating the discoverable document with the semantic context.