Sunday, May 3, 2026
Saturday's Build Your Own LLM workshop was a success. Eight hours of focused instruction that took us from a single perceptron all the way through evals, instruction fine-tuning, and reinforcement learning. Everyone in the room walked out with the concepts and math required to train GPT-2 and a real picture of how each layer fits together under the hood.
Huge thanks to Justin Angel for teaching the session. He built the full curriculum solo and the depth came through. We're working on a YouTube version next so the people who couldn't make it can still go end-to-end.
Saturday won't be the only chance, and we're working on more.
Community Spotlight
Abhi Gadudasu
Abhi built Novi Books, a book discovery app that learns your taste from head-to-head book rankings. Check it out in the Apple app store.
Coming up at Rally
AI Benchmarks for Human Flourishing
2026-05-04 · 6:45 PM - 8:45 PM
We benchmark AI for accuracy, speed, and reasoning. HumaneBench tests whether the model is actually good for the person using it. Erika Anderson and Yaoli Mao built it on care ethics, tested it in production, and found that model behavior flips 67% of the time when users are vulnerable. Talk plus a working session on building one for your own product.
Builders Night
2026-05-11 · 6:30 PM - 9:00 PM
Monday co-working for AI builders. Two hours of focused build, then a share-back round and demos to close.
AI Research Circle: Physical AI
2026-05-25 · 7:00 PM - 8:45 PM
Atharva Atre on physical AI and the shift from software-only systems to AI that touches the physical world. Conversational seminar on training approaches, the industry landscape, and where the field is heading.
Around SF this week
Wednesday, May 6, 5:30 PM - 7 PM
**Build Your First Agent (ft. Hermes)** Get hands-on with Hermes and ship a working AI agent in one session. You'll learn the architecture decisions that matter and walk away with code you can actually use. Perfect if you've been meaning to build with agents but haven't carved out the time.
Thursday, May 7, 5:30 PM - 8:30 PM
The 2026 Americas Catan Champion will break down strategy, probability, and decision-making—then we'll explore how those same concepts apply to AI systems. You'll play a few rounds, discuss trade-offs in both games and models, and leave with a sharper intuition for multi-agent dynamics. Hosted at Sphere, SF's new tech/gaming space in SoMa.
Saturday, May 9, 11 AM - 5:30 PM
Build AI interfaces that actually generate themselves. This hackathon focuses on agentic UI patterns — apps that use A2UI, AG-UI, and Model Context Protocol to let agents construct their own interfaces on the fly. You'll leave with working code and a sharper sense of where generative UIs beat static ones.
Saturday, May 9, 11 AM - 8 PM
# Autonomous GTM Hackathon Build AI agents that actually handle go-to-market work — from lead research to outbound sequences to demo scheduling. You'll ship working prototypes alongside other builders tackling the same problems, with feedback from founders who've automated parts of their sales motion. Bring your laptop and an API that does something useful.
Wednesday, May 13, 6 PM - 9 PM
Fin (Intercom's AI agent) and Metronome (usage-based billing infrastructure) are teaming up to tackle the thorniest problem in AI products: how to price agent work when costs and value are both hard to predict. You'll hear real pricing models from companies shipping agents in production, plus the billing architecture that makes complex usage pricing actually work. If you're figuring out whether to charge per conversation, per resolution, or per API call, this is your crash course.