Berlin Builder Circle #3: Small Models in Production
Small models are moving from demos into production. Teams post-train them to replace the frontier-model calls their agents make thousands of times a day, and ship them on-device for privacy, offline use and cost. The hard parts rarely make it into talks: when it's worth it, which data and evals you can trust, what breaks after launch, and how runtimes and hardware behave on real devices.
Berlin Builder Circle #3 is a small, closed-door session for practitioners working on exactly that. It follows our Small Language Models Night in October: the talks raised the questions, and this is where builders dig into them.
No talks, no slides, no vendor pitches. Everyone contributes.
What we'll cover
When a post-trained small model beats prompting a frontier model, and when it doesn't
Finding the step worth replacing in your production traces
Synthetic data, evals, and keeping quality up after launch
On-device runtimes, hardware limits, and shipping model updates
The questions you send in when you apply
Agenda
18:00 Doors, food and drinks
18:20 Welcome from the hosts and Rasa
18:30 Round-robin: what you've shipped, where you're stuck, what you want from the room (2 minutes each)
19:15 Short break
19:25 Guided discussion, part 1: when is a small model worth it?
19:55 Guided discussion, part 2: shipping on-device
20:20 Wrap-up
20:30 Close
Who should apply
Engineers, researchers and technical leads who have post-trained, served or shipped small or fine-tuned models in production, or are building one for real users right now. Mobile and edge developers shipping AI on-device are very welcome.
Entry requirement: you've worked on the topic yourself. If you're not actively building in this space, this probably isn't the right room; our public AAIF Berlin meetups are open to everyone.
About 20 seats. We read every application and approve on a rolling basis until Friday 6 November. Please be specific about what you're building: it's how we put the room together and choose the discussion questions.
We use the Chatham House Rule: share what you learn, not who said it.
Hosted at the Rasa office. Thanks to the Rasa team for having us again.
About AAIF
The Agentic AI Foundation (AAIF) is a community dedicated to advancing the understanding and practical application of agentic AI. We bring together engineers, researchers, founders, builders, and AI enthusiasts to explore how autonomous AI systems are designed, evaluated, and deployed.
Through reading groups, workshops, and community discussions, AAIF brings together engineers, researchers, founders, and AI enthusiasts to exchange ideas, challenge assumptions, and learn from one another.
By attending you agree to our Code of Conduct and Privacy Policy.