GenAI
Claims Evidence Agent (RAG)
Retrieve policy docs + generate citation-backed summaries for denial workflows.
RAGPythonVector SearchDatabricks
Overview
A retrieval-augmented assistant that finds supporting policy/claim documents and generates short, citation-backed summaries to speed up denial review workflows.
Problem
- Reviewers spend time searching across scattered policy documents.
- Summaries must be defensible and traceable to source evidence.
Solution
- Index documents into a vector store for semantic retrieval.
- Use RAG to generate reviewer-friendly summaries with citations.
- Log prompts, sources, and outcomes to support audits.
Architecture
- Documents → chunking → embedding → vector index
- Query → retrieve top-k evidence
- LLM summary generation with citations
- Output stored for reviewer workflows
Metrics
- Reduced time-to-evidence for common denial cases.
- Higher consistency in reviewer notes.
- Citation coverage tracked per response.
Highlights
- Citation-first outputs for trust.
- Designed for enterprise guardrails and auditing.
- Fits naturally into existing data platforms.