An evening where open agents meet open tabular models. This is a room for the people who build agents they own end-to-end: AI engineers building agentic systems and ML engineers building the models underneath them, together, at deepset's HQ in Berlin.Powered by Haystack (open-source AI orchestration framework) and Prior Labs (creators of TabPFN, an open-weight foundation model for tabular data), the sessions are shaped by what you're actually building, your experience, and your challenges in assembling open agent stacks.

🎯 Open Weights, Open Agents

Agents have stopped being a single LLM that does everything. They now run inside a harness that manages the loop of reasoning and acting, decides which tool or model to call next, maintains memory and state across steps, and adds the observability and guardrails that make an agent trustworthy enough to ship. Haystack's orchestration turns a model into a reliable system and lets you compose specialized agents that hand off work to models best suited to the subtasks they're actually good at.

One of those subtasks is prediction on structured data: the kind of tabular problem that generative models are a poor fit for and where classical ML still wins. Prior Labs' TabPFN is a foundation model purpose-built for exactly that: trained once, it makes strong predictions on new tabular datasets without task-specific fine-tuning, and it ships as open weights, so you retain control of your agent all the way down to the models it calls.

Some example topics we'll discuss:

  • Why you need orchestration for your agent

  • Working with structured data in agentic systems

  • Agents that need to make grounded predictions

  • Benchmarking for tabular models: why LLM benchmarks don't apply, and why this gap exists

  • Agents and tool selection: when agents should hand off to TabPFN for prediction tasks

  • What "owning your stack end to end" actually requires in practice

  • Open-weight vs. closed models: what changes when you can inspect, fine-tune, or self-host the thing

👥 Who is This Event For?

This event is for builders and decision makers on both sides of the agent stack:

  • AI / agent engineers building agentic workflows running in production

  • ML engineers and data scientists working on structured data, tabular models, and prediction tasks

  • Product managers and tech leads

  • Founders and engineering managers

  • Innovation leads and decision makers exploring AI adoption

The goal is to share experiences, ask questions, and learn from others who are building or enabling real-world AI systems.

🧠 Unconference Event Format

This event follows the unconference format. We’ll form discussion groups based on topics you shared during registration, so the agenda reflects what matters most to you.

Bring your questions, share your learnings, or simply listen in and learn from others in the field.

​🗓️ Agenda

18:00 - 18:30 Arrival & Welcome Drinks 🥙🍹18:30 – 18:45 Opening Remarks18:45 – 19:45 Unconference Sessions19:45 – 20:30 Wrap-Up & Networking

Use the "Contact the Host" feature on Luma if you have any questions!

📸 Some Moments From Previous Unconferences