Documentation Index
Fetch the complete documentation index at: https://mintlify.com/avnlp/agentic-med-diag/llms.txt
Use this file to discover all available pages before exploring further.
This page covers every step needed to go from a bare machine to a fully configured Agentic Graph RAG environment: installing the Python dependency manager, syncing the project lockfile, spinning up the Neo4j and Milvus storage backends in Docker, and setting the environment variables the pipeline needs to connect everything together.
Requirements
Before you begin, confirm that the following are available on your system:
| Requirement | Purpose | Notes |
|---|
| Python 3.10+ | Runtime for all pipeline and evaluation code | Check with python --version |
| uv | Fast Python package and project manager | Installed in the next section |
| Docker | Runs Neo4j and Milvus as local containers | Any recent stable release |
| LLM API access | Powers the reasoning steps inside the pipeline | e.g., an OpenAI API key |
Installing uv
The project uses uv for dependency management. Install it with pip:
uv manages a project-local virtual environment automatically. You do not need to create or activate a venv manually — running uv sync is sufficient.
Installing Project Dependencies
Clone the repository and let uv install all locked dependencies in one step:
git clone https://github.com/avnlp/agentic-med-diag.git
cd agentic-med-diag
uv sync
uv sync reads the project lockfile and installs all dependencies into a .venv directory inside the project root. Subsequent uv run calls automatically use that environment.
Neo4j Setup
Neo4j stores the entity/relationship knowledge graph built from ingested medical documents. The relational retrieval channel queries Neo4j to fetch structured KG triples and multi-hop paths during inference.
Start a local Neo4j instance with Docker:
docker run --name neo4j \
-p 7474:7474 -p 7687:7687 \
-e NEO4J_AUTH=neo4j/password \
neo4j:latest
| Port | Protocol | Purpose |
|---|
7474 | HTTP | Neo4j Browser UI |
7687 | Bolt | Driver connection used by the pipeline |
neo4j / password (as set by NEO4J_AUTH above).
For production or persistent development use, mount a volume (-v $PWD/neo4j-data:/data) so graph data survives container restarts.
Milvus Setup
Milvus stores dense vector embeddings for semantic retrieval over document chunks. The semantic retrieval channel performs hybrid search over Milvus to ground sub-queries in text evidence during each reasoning hop.
Start a standalone Milvus instance with Docker:
docker run -d --name milvus-standalone \
-p 19530:19530 -p 9091:9091 \
milvusdb/milvus:latest \
milvus run standalone
| Port | Protocol | Purpose |
|---|
19530 | gRPC | Primary connection used by the pipeline |
9091 | HTTP | Metrics and health endpoint |
For persistent development, mount a volume (-v $PWD/milvus-data:/var/lib/milvus) to retain indexed embeddings across restarts.
Environment Variables
The pipeline reads connection details and credentials from environment variables. Set these before running any pipeline code:
| Name | Description |
|---|
NEO4J_URI | Bolt URI for the Neo4j instance, e.g. bolt://localhost:7687 |
NEO4J_USERNAME | Neo4j username, e.g. neo4j |
NEO4J_PASSWORD | Neo4j password, e.g. password |
MILVUS_HOST | Hostname or IP of the Milvus instance, e.g. localhost |
MILVUS_PORT | gRPC port for Milvus, e.g. 19530 |
OPENAI_API_KEY | API key for your LLM provider (OpenAI or compatible) |
export NEO4J_URI="bolt://localhost:7687"
export NEO4J_USERNAME="neo4j"
export NEO4J_PASSWORD="password"
export MILVUS_HOST="localhost"
export MILVUS_PORT="19530"
export OPENAI_API_KEY="sk-..."
.env file at the project root — uv run will pick them up automatically.
Verifying the Installation
After completing the steps above, confirm that both services are healthy:
# Check Neo4j is reachable on the Bolt port
nc -zv localhost 7687
# Check Milvus health endpoint
curl http://localhost:9091/healthz
OK. For Neo4j, open http://localhost:7474 in a browser and confirm you can log in.
The codebase is actively being developed and the full pipeline code will be released soon. Once released, you can verify the Python installation by importing the top-level package after running uv sync.
Do not expose the Neo4j or Milvus ports publicly without additional authentication and network controls. The Docker commands above are intended for local development only.
