RAISE Summit AI - Mentoring and Judging the Hackathon in Paris

There's a particular kind of energy in a room full of people who have a weekend, a pile of GPUs, and something to prove. 

That was Paris this month, and I got to spend it as a mentor and judge at the RAISE Summit AI Hackathon.

Let me back up a bit, because this was my first hackathon wearing a SUSE hat, and it turned into exactly the kind of hands-on event I signed up for when I joined as Technical Advocacy Lead.

Let's look at how the attendees kicked this weekend off.

The pitch behind SUSE AI Factory with NVIDIA is going from do-it-yourself AI to something you can actually run and maintain: a sovereign, secure platform where Rancher is the single control plane for VMs, containers, and AI workloads, and validated blueprints handle the workload-specific plumbing for things like GenAI inference, RAG, and MLOps. That's the enterprise story, and it's a good one.

These are the slides presented in the opening keynote.

Getting started with SUSE stack

But a hackathon doesn't care about your enterprise deployment architecture diagram. A hackathon cares about one thing: can I get something running in the next ten minutes so I can start building?

So rather than walk teams through slides, I gave them a stack they could stand up on their own laptops. Everything lives in a single project: rancher-hackathon-paris. Clone it, run one script, give it your hackathon registration email, and you're off:

# The repository is only available during the event.

#

$ git clone git@gitlab.com:eschabell/rancher-hackathon-paris.git

$ cd rancher-hackathon-paris

$ ./init.sh (or init.bat for windows users)

That init.sh does the unglamorous work nobody wants to do at 9am on day one: it checks prerequisites, starts Rancher Desktop headlessly, waits for the k3s node to become ready, logs Docker and Helm into the registry, drops the image pull secret into the hackathon namespace, installs the Application Collection extension, and writes out an MCP configuration. By the time it prints "Setup complete," you have a working Kubernetes cluster and a catalog of production-grade software a browser-click away.

The AI developer stack

The SUSE Application Collection is 148 applications of trusted, secure open source, and for an AI hackathon the relevant slice is deep:

• 
  AI/ML: ollama, vLLM, Open WebUI (plus pipelines and MCP-over-OpenAPI), PyTorch, Milvus, MLflow 
• Data and infra: PostgreSQL, MariaDB, Redis/Valkey, MinIO, OpenSearch, NATS, Kafka
• Platform: cert-manager, Keycloak, Istio, Harbor, Prometheus, Argo CD, Fluent Bit

The moment that seemed to interest attendees was the MCP server. 

Hand the generated mcp.json to a tool like Cursor and the assistant can see the entire Application Collection and help deploy from it. Watching a team go from "what's in here?" to a deployed vector database by talking to their editor — that's the kind of thing that makes people stop treating the platform as overhead and start treating it as leverage. If a tool rejects basic auth, there's a local proxy (./init.sh --mcp-proxy), and the secret ships to any namespace with ./init.sh --deploy-secret <namespace>. Small touches, but they're the difference between a demo and a thing people actually use.

Mentoring and judging

The most fun was to wander around, engaging with the 5 person teams trying to make their ideas a reality. I don't want to share the ideas I've seen already in the first hour (I'm writing live from the hackathon floor!), but this is going to crazy to see how much is achievable with AI and bright minds.

The prize

The winning team will earn an array of prizes, one of which are SUSE eLearning vouchers — five Silver licenses, 90 days of full access across the SUSE portfolio, at suse.com/elearning. This is a great hackathon prize where a weekend gets you a prototype, but structured learning will provide you the ability to build the next ten. That's the right thing to be handing the people who already showed they'll do the work.

Conclusion

Hackathons are where we get to test whether our "industrialized AI" story survives contact with people who just want to ship. In Paris, the stack is being used, the Application Collection is ready to do the heavy lifting, and the teams are finding out why this work is fun in the first place.

If you were there and want to keep the conversation going, or you're building something interesting on top of Rancher and the Application Collection, find me on the usual channels. Let's build something worth talking about.