Policy Intelligence

## Inspiration

Drug coverage policy is fragmented across payer PDFs, formularies, and spreadsheets, which makes basic analysis slow and inconsistent. We built Policy Intelligence to give analysts one place to ingest, normalize, query, and compare coverage rules with confidence.

## What it does

Policy Intelligence ingests payer artifacts, converts them into a common plan-drug-rule structure, and helps users answer practical questions like:

• Which plans cover a drug?
• What prior auth criteria apply?
• How do plans differ?
• What changed over time?

It combines deterministic comparison with grounded AI explanations so users can move faster without losing trust.

## How we built it

We built the product with:

• React for the frontend
• Express for APIs and ingestion
• Ollama for local AI responses
• a normalization layer for plans, drugs, and rules
• deterministic query and compare functions over structured data

The AI layer explains results, but the underlying answers come from structured records and tool outputs.

## Challenges we ran into

The biggest challenge was correctness. Payer data comes in inconsistent formats, especially PDFs, so normalization and trust were much harder than simple ingestion. We also had to make sure the AI stayed grounded in structured results rather than generating unsupported answers.

## Accomplishments that we're proud of

We built more than a chatbot. We created a workflow that supports ingestion, normalization, query, and comparison in one system, with provenance, confidence signals, and analyst-friendly compare views.

## What we learned

We learned that in this space, structured data and normalization matter more than flashy AI. The strongest pattern is deterministic retrieval first, AI explanation second.

## What's next for Policy Intelligence

Next, we want to improve policy change tracking, deepen PDF normalization, add richer provenance views, and make the compare experience even stronger for real analyst workflows.

Built With

• csv
• data
• express.js
• file-based
• json
• local
• node.js
• ollama
• pdf-parsing
• policy-normalization
• react
• sse-streaming
• typescript
• vite