Skip to content
All work
Full Stack2026

OPD Claims Adjudication Engine

An AI insurance-claims processor that reads uploaded medical bills and adjudicates them against policy rules.

FastAPIMongoDBGemini VisionLangChainFAISSNext.js
Source

The problem

Outpatient claims arrive as photographs of paper. Getting from a blurry bill to an approve/reject decision is mostly extraction and rule-matching, and both are tedious enough to automate.

How it works

A Gemini vision model extracts structured line items from uploaded bill images, which are stored in MongoDB alongside the claim record.

Policy rules are retrieved with a LangChain/FAISS RAG layer, so adjudication reasons against the specific clauses that apply to a claim rather than a generic prompt, and returns a decision with the clause it relied on.

A Next.js front-end covers submission and review; the FastAPI backend holds the adjudication logic.

Outcomes

  • Image-to-decision in one flow: extraction, retrieval, adjudication.
  • Every decision traceable to the policy clause behind it.
  • Built as the AI Automation Engineer assignment for Plum.