HAICON26 Agentic RAG Hackathon

Hands-on mini hackathon for the Helmholtz Agentic AI Workshop at HAICON26.

Build, plan, or stress-test a minimal PDF research bot: MCP tools + four steps (Discover → Select → Read → Answer) — a simple RAG pipeline over local files.

Workshop repo (slides & overview): Helmholtz-AI-Matter/agentic-ai-workshop

Also at the hackathon: Abstracts Explorer — choose one project for the session.

Setup

git clone https://github.com/haider-khan-91/haicon26-agentic-rag-hackathon.git
cd haicon26-agentic-rag-hackathon
conda env create -f environment.yml
conda activate hackathon-haicon
cp .env.example .env   # optional: OPENAI_API_KEY for full LLM run
python scripts/generate_sample_pdfs.py

.env must live in the project root (same folder as run_bot.py), not in agent/ or a parent folder.

If you change .env, run the command again (each python run_bot.py is a fresh process).

If the key still isn’t picked up: check you’re not overriding it in the shell (echo $OPENAI_API_KEY). The bot loads .env with override enabled.

For a custom API endpoint (Azure, proxy, etc.), set OPENAI_BASE_URL in .env — include the /v1 path, e.g. https://your-resource.cognitiveservices.azure.com/openai/v1/. Set OPENAI_MODEL to your deployment name.

Run all commands from the project root (the folder containing run_bot.py). Otherwise you may see ModuleNotFoundError: No module named 'agent'.

Run

python run_bot.py --dry-run
python run_bot.py "What methods are used?"   # needs API key in .env
python scripts/integration_demo.py           # Track C, no API key

Add your own PDFs to papers/.

Hackathon tracks (pick one)

See PARTICIPANT_SHEET.md and TRACKS.md.

Track	Focus	File
A Feature sprint	Extend the bot (new MCP tool or pipeline step)	FEATURE_BACKLOG_Track_A.md
B Architecture planning	Plan agentic integration for your or another project — no code	DESIGN_PROPOSAL_Track_B.md
C Integration experiments	Run tests, compare tools vs bot, report failures	INTEGRATION_Track_C.md

Fill in the deliverable section at the bottom of your track file before share-out.

Layout

• mcp_servers/ — PDF and system MCP tools
• agent/ — MCP client and research bot
• run_bot.py — CLI entry point