# Multi Document Upload Service

This service accepts large batches of heterogeneous documents, extracts causal
relationships with Claude Sonnet 3.5, and writes them into Neo4j as a
knowledge graph.

## Features

- Multipart upload endpoint (`POST /jobs`) capable of handling dozens of files
  and mixed formats (PDF, DOCX, PPTX, XLSX/CSV, JSON/XML, images, audio/video).
- Content extraction powered by the `unstructured` library with fallbacks.
- Chunking tuned for Claude Sonnet (800 token target, 200 overlap).
- High-accuracy causal extraction using Anthropic Claude with provenance.
- Neo4j graph writer that upserts `Concept` nodes and `CAUSES` edges.
- Status endpoint (`GET /jobs/{id}`) and graph summary endpoint
  (`GET /jobs/{id}/graph`).

## Configuration

Environment variables:

- `ANTHROPIC_API_KEY` (required)
- `MULTI_DOC_CLAUDE_MODEL` (default `claude-3-5-sonnet-20241022`)
- `NEO4J_URI` (default `bolt://localhost:7687`)
- `NEO4J_USER` / `NEO4J_PASSWORD` (default `neo4j` / `neo4j`)
- `MULTI_DOC_STORAGE_ROOT` (default `storage` inside project)

## Run locally

```bash
uvicorn multi_document_upload_service.main:app --reload --host 0.0.0.0 --port 8035
```

Ensure Neo4j is reachable and Anthropic credentials are exported before
starting the service.