README.md

---
title: Payer AI Prototypes
emoji: 🚀
colorFrom: indigo
colorTo: pink
sdk: docker
pinned: false
---

Payer AI Prototypes

# Payer AI Prototypes

Two quick healthcare payer prototypes in one Chainlit app:

1. **Claims Ingestion & Normalization**
2. **Automated Scheduling & Admin Assistant**

The app uses:

- `uv` for dependency management
- Chainlit for UI
- LangGraph for deterministic agent workflows
- LangChain tools
- in-memory Qdrant for prototype RAG retrieval
- Loguru for logging

## Folder layout

```text
payer-ai-prototypes/
  app.py
  graphs.py
  agents.py
  tools.py
  schemas.py
  prompts.py
  config.py
  data/
    claims/
      mock_claim_packets/
      rag/
        policy_benefit_rag/
        exception_similarity_rag/
    scheduling/
      mock_data/
      rag/
        provider_specialty_matching/
  scripts/
    ingest_claims_rag.py
    ingest_scheduling_rag.py
  tests/
  pyproject.toml
  Dockerfile
  .env.example
```

## Where to place your already-downloaded mock data

### Prototype 1: Claims Ingestion & Normalization

Place the unzipped `mock_claim_packets.zip` contents here:

```text
data/claims/mock_claim_packets/
```

Expected example shape:

```text
data/claims/mock_claim_packets/
  clm_001/
    claim.json
    provider_note.txt
    attachment.pdf
    fax_scan.png
  clm_002/
  ...
```

Place the unzipped `claims_rag_datasets.zip` contents here:

```text
data/claims/rag/
  policy_benefit_rag/
    policy_benefit_chunks.jsonl
    policy_benefit_metadata.csv
    source_docs/
  exception_similarity_rag/
    resolved_exception_cases.jsonl
    exception_similarity_metadata.csv
    case_files/
```

### Prototype 2: Scheduling & Admin Assistant

Place the scheduling mock data files here:

```text
data/scheduling/mock_data/
  members.csv
  benefits.csv
  referrals.csv
  authorizations.csv
  provider_availability.csv
  specialist_locations.csv
```

Place the provider/specialty matching RAG dataset here:

```text
data/scheduling/rag/provider_specialty_matching/
  provider_specialty_profiles.jsonl
  provider_specialty_metadata.csv
  source_docs/
```

If your file names are slightly different, update the paths in `config.py`.

## Local setup

```bash
uv sync
cp .env.example .env
uv run chainlit run app.py
```

## Optional RAG ingestion

The tools auto-load JSONL files into in-memory Qdrant at runtime. You can also explicitly test ingestion:

```bash
uv run python scripts/ingest_claims_rag.py
uv run python scripts/ingest_scheduling_rag.py
```

## Hugging Face Spaces deployment

Create a Docker Space and push this repo. Add secrets:

- `OPENAI_API_KEY`
- optionally `HF_FT_EMBED_MODEL_URL`

HF Space command is handled by the Dockerfile:

```bash
uv run chainlit run app.py --host 0.0.0.0 --port 7860
```