Welcome to The Agent Hackathon!
Everyone's building AI agents, but the real magic happens when those agents have complete context to act on organizational data. Without reliable knowledge of schemas, lineage, ownership, ML metadata, and governance, agents hallucinate or get stuck on tasks any data engineer could finish in minutes. That's where DataHub comes in. DataHub is the open-source context platform that gives AI agents a complete understanding of your data stack — from raw tables to ML models. With an MCP Server, end-to-end ML lineage, and DataHub Skills that give agents direct access to catalog workflows, DataHub turns the modern data ecosystem into something agents can actually work with.
This hackathon is your invitation to build on that foundation. Whether you're shipping autonomous agents, generating production data code, protecting ML models, or building something entirely new — show us what's possible when agents have context.
What to Build
Create a working software application that uses DataHub to solve one of the challenges below. Pick one of the four challenges (or combine them):
- Agents That Do Real Work: Build AI agents that handle data problems — alone or as a team. Your agent reads DataHub through the MCP Server or Agent Context Kit to understand what's connected to what, takes action, and writes results back so the next person or agent inherits the knowledge.
- Metadata-Aware Code Generation & Development: Build agents that generate production data code — transformation models, pipeline DAGs (Airflow, Prefect, Dagster), ingestion scripts, helper scripts, configurations, migration code — that works on the first try because they use DataHub Skills or the MCP Server to read DataHub for the real schemas, lineage, and rules before generating anything.
- Production ML Agents: Build agents for ML teams that protect models in production. Use DataHub's end-to-end ML lineage — the path from training data to features to models to deployments — accessed via the Agent Context Kit or MCP Server to catch silent problems that can break ML systems before they cost money.
- Open / Wildcard: Build anything creative that uses DataHub as the foundation — supply chain optimization, financial forecasting, regulatory automation, knowledge capture, or anything else.
What to Submit
- Include a URL to your Project that gives judges easy access to test the functionality — a live demo, hosted app, or your repo with clear setup instructions.
- Provide a URL to your public code repository for judging and testing. The repository must contain all necessary source code, assets, and full instructions required for the project to be functional. The repository must be public and open source by including an Apache 2.0 open source license file.
- Include a text description that summarizes your Project that might include describing its features, functionality, technologies, and data you used.
- Include a demonstration video of your Project that is under 3 minutes, uploaded to YouTube or Vimeo with public visibility enabled.
- Optional: Include Sample outputs.
Prizes
- Grand Prize: $6,000, Presentation at DataHub Townhall, Social media @ Slack community promotion, Special LinkedIn Badge.
- Challenge Winners: $3,000, Social media @ Slack community promotion, Special LinkedIn Badge.
- Honourable Mention: $1,000, Special LinkedIn Badge.
- Most Valuable Feedback Survey Prize: $50 × 10 awards.
Judging Criteria
- Use of DataHub: How meaningfully does the project use DataHub — its context graph, MCP Server, Agent Context Kit, DataHub Skills, or Analytics Agent?
- Technical Execution: Quality of implementation, robustness, and whether the project actually works end-to-end.
- Originality: How creative and novel is the approach?
- Real-World Usefulness: Would a real data, ML, or AI platform team see clear value in this?
- Submission Quality: Quality of the demo video, written description, and README.
- Bonus: Meaningful Open-Source Contribution: Submissions that include meaningful open-source contributions to DataHub will be looked on favorably.