Skip to content

Add RAG + Validation upgrade proposal (documentation)#7

Open
delisha02 wants to merge 2 commits intomainfrom
codex/clarify-fact-extraction-processes-and-algorithms
Open

Add RAG + Validation upgrade proposal (documentation)#7
delisha02 wants to merge 2 commits intomainfrom
codex/clarify-fact-extraction-processes-and-algorithms

Conversation

@delisha02
Copy link
Owner

Motivation

  • Provide a roadmap to upgrade the drafting pipeline to a legally grounded, auditable, and citation-aware system with higher practical quality.
  • Address current gaps including missing retrieval in generation, isolated research RAG, lack of clause-level tagging/traceability, and limited validation capabilities.
  • Establish a clear default of non-agentic RAG for routine drafting with a gated agentic mode for complex cases to balance latency, cost, and determinism.

Description

  • Adds docs/rag_upgrade_proposal.md which specifies a recommended target architecture, including hybrid retrieval, clause-by-clause generation, a legal validation engine, citation traceability, and export formats.
  • Defines priority workstreams P0/P1/P2 with concrete technical changes such as integrating retrieval into generate_document, implementing get_hybrid_retriever(...), adding clause schemas and per-clause provenance, and a deterministic validator with LLM-assisted fallbacks.
  • Lists API and data-model additions including response fields for content, citations[], validation_report, clause_trace[], and persistent JSON fields like retrieval_sources.
  • Includes rollout strategy, evaluation metrics to track (e.g., citation precision@k, missing clause rate, hallucination rate), risk controls, and minimum controls before full agentic rollout.

Testing

  • This is a documentation-only change and no automated tests were run for this PR.

Codex Task

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant