​If you can write a prompt, you can train a model.

​Join distil labs, langfuse, and runpod to kick off your GTC Berlin week with a hackathon. We're bringing together a stellar group of builders to train their custom small models and build an improvement loop from production traces.

​What to expect

​First things first, there's going to be free pizza, drinks and a 1000€ cash prize 🔥🔥🔥 Bring a laptop and your own use case, or start with one of our prepared examples. Working in small teams, you'll:

• ​Find the weak spots. Use Langfuse traces and evaluations to understand how your agent behaves on real requests.

• ​Train and improve a small model. Turn those traces into evaluation and training data with distil labs, then use autoresearch to train, inspect mistakes, and iterate toward your quality target within a set budget.

• ​Deploy and repeat. Serve your trained model on Runpod, capture its new interactions in Langfuse, and use what you learn to guide the next version.

​The distil labs engineers will be there to help you work through the loop.

​Who should join

​AI engineers, software developers, ML engineers, and technical founders working on agents or LLM applications. Builders who are comfortable using Claude Code or similar coding agents to build and test applications are welcome, too.

​No previous fine-tuning experience is required, and we’ll have prepared examples if you don’t have an agent of your own yet.

​About the hosts

​distil labs is the post-training platform where engineers build high-performance agents and continually improve them using production data.

​Langfuse is an open-source platform where engineers trace, evaluate, and improve AI agents.

​Runpod is a GPU cloud platform where developers train, deploy, and scale AI models.