Inference-Time Compute Hackathon
- Date
- 2026-06-20
- Location
- San Francisco, CA, USA
- Host
- Etched
About this event
Inference-time compute is becoming one of the most important levers in modern AI: not just how a model is trained, but how intelligently it thinks, searches, verifies, and adapts at the moment you ask it to do real work. This hackathon is built for people who want to explore that frontier hands-on, with others who care about technical depth, practical experimentation, and the future of model performance. If you have been thinking about routing, tool use, multi-step reasoning, test-time scaling, agent loops, verification, or any system that trades extra compute at inference for better outcomes, this is the room you want to be in. Come to build, compare ideas, and pressure-test what actually works. About the Event The Inference-Time Compute Hackathon is an in-person gathering in San Francisco centered on building and sharing projects that use compute at inference time in thoughtful ways. Rather than focusing on model training or abstract discussion alone, the event is designed around implementation: taking an idea, turning it into a prototype, and learning from other builders doing the same. This is a hackathon, but it is also a community event. Expect a mix of independent hacking, technical conversations, spontaneous collaboration, and networking with people who are actively working on adjacent problems. Whether your interest is research-oriented, product-oriented, or somewhere in between, the format supports both focused building and useful cross-pollination. The theme is intentionally broad enough to invite creativity while staying technically grounded. Projects might explore better reasoning pipelines, evaluation-informed generation, retrieval and reranking strategies, self-critique loops, adaptive compute allocation, orchestration frameworks, or entirely new approaches to getting more capability out of models at run time. What to Expect The event begins on Friday, June 19 at 5:00 PM PDT and is structured to help people get productive quickly. You can expect a practical, builder-friendly environment where people arrive with ideas at different stages: some with a clear project plan, others looking for a teammate, a problem to tackle, or a sharper framing for something they have already been exploring. A typical flow for an event like this includes: Arrival and check-in with time to meet other attendees Project idea sharing for people who want to pitch a concept or find collaborators Focused hacking time to build prototypes, experiments, demos, or evaluation setups Technical discussion and feedback as teams compare approaches and troubleshoot in real time Informal networking with engineers, researchers, founders, and curious practitioners You should also expect a lot of conversation around tradeoffs, not just demos. Inference-time compute is full of practical questions: when extra steps help, when they do not, how to measure gains, how latency affects usefulness, and what makes a technique robust outside a toy example. That makes this a strong setting for people who like both building and critical thinking. Because the event is in person, one of the biggest advantages is the speed of iteration you get from being in the same room. You can sketch an idea, get immediate reactions, pair on implementation details, and refine your approach without the friction of fully remote coordination. Why Attend This event is valuable if you want to move beyond general AI enthusiasm and into a more specific, high-signal area of experimentation. Inference-time compute is where a lot of important work is happening right now, and a hackathon format gives you a fast way to learn by doing rather than just reading about techniques after the fact. You will leave with more than a loose sense of the topic. Depending on how you participate, you may come away with a working prototype, a clearer technical thesis, new collaborators, better intuitions about system design, or a stronger understanding of what kinds of test-time strategies are actually promising in practice. There is also real value in the room itself. Events like this attract people who care about the mechanics of AI systems, not just the headlines. If you are looking to meet others who think carefully about model behavior, evaluations, orchestration, and applied research questions, this is a highly relevant crowd. A few concrete reasons to attend: Build something ambitious in a focused setting Get feedback from technically curious peers Meet collaborators for future projects or research directions Explore a fast-moving topic through direct experimentation Pressure-test ideas against real implementation constraints Practical Details This event takes place in person in San Francisco, USA, which makes it a good fit for attendees who want face-to-face collaboration and live technical exchange. If you do your best work around whiteboards, laptops, quick conversations, and side-by-side debugging, the format will suit you well. The listed start time is Friday, June 19 at 5:00 PM PDT. Plan to arrive on time so you can settle in, meet other attendees early, and take part in any opening coordination or idea-sharing. Early conversations often shape teams, projects, and momentum for the rest of the event. A few ways to prepare: Come with a project idea, even if it is still rough Or come ready to join someone else's idea if you prefer collaboration Bring your laptop and anything you need to prototype comfortably Be ready to explain what you want to build in a few clear sentences Think about how you would evaluate success, not just what you want to make If you are excited by the question of how to get better performance, reliability, or reasoning out of models at inference time, this hackathon offers the right combination of technical focus and community energy. It is a place to test ideas seriously, meet people who care about the same problems, and spend your Friday evening building at the edge of what AI systems can do.
Who should attend
This is for you if you want to explore AI systems through building, experimentation, and sharp technical conversation. - **You are an engineer or researcher** interested in reasoning systems, agents, evaluation, retrieval, orchestration, or test-time scaling, and you want a hands-on setting to push ideas forward. - **You have a prototype or half-formed concept** around inference-time compute and want feedback, collaborators, or the pressure of a real room to turn it into something concrete. - **You care about practical AI performance**, not just model selection, and you are curious about when extra inference-time steps actually improve quality, reliability, or usefulness. - **You like hackathons that attract technical peers** and want to meet people in San Francisco working on adjacent problems across research, product, and infrastructure. - **You are a founder, builder, or advanced practitioner** looking for signal on a fast-moving area that could shape the next generation of AI products and workflows. - **You enjoy learning by making**, comparing approaches, and discussing tradeoffs like latency, verification, cost, and evaluation with others who take those questions seriously.