Open Source Computer Use Hackathon | Nvidia, Tzafon, SF Compute, Kernel

​We’re inviting an exclusive group of local builders to spend a day at the first computer use hackathon in San Francisco.

​Join us to build cool projects with open source AI & computer use. It can be anything from fine-tuning models to improved verification and data collection to full end-to-end projects.

​Throughout the hackathon, you will get unlimited access to both KERNEL’s open source infrastructure and Tzafon’s computer use model series.

Schedule

• ​9:30am: Doors open (please arrive on time before 10 am)

• ​10:00am: Intro panel on the state of computer use

  • ​Marius Buleandra, Member of Technical Staff, Anthropic

  • ​Simon Koser, Co-founder, Tzafon

  • ​Catherine Jue, CEO & Co-founder, KERNEL

• ​10:30am: Start hacking

• ​12:00pm: Lunch served (🍕 Pizza / 🥬 Salads available)

• ​4:00pm: Final submissions

• ​4:30pm: Winner presented

• ​4:45pm: Post-event social mixer

​Prizes

• ​1st place: $10k + $10k in credits

• ​2nd-5th place: $1k + $5k in credits

​Teams

​Teams up to 5. Solo applicants will be pre-matched with a team.

​Signup

​Because we have limited space we're asking that you change your RSVP if anything changes after you've signed up.

​If you are applying as a team, please add all your teammates when signing up (group ticket).

​Judge Panel & Tracks

• ​Dual track: CUA applications and CUA model research

  • ​CUA applications: apply computer use to some cool domain, e.g. using KERNEL for browser infra & Northstar for computer models

  • ​CUA model research: more oriented around RL, fine-tuning, vision encoders, evals, etc. Nvidia provides compute credits during event.

• ​Judge criteria: adherence to track, impact, technical novelty, presentation, and whether it contributes to open source.

​The judge panel consists of:

• ​Catherine, co-founder KERNEL

• ​Simon, co-founder Tzafon

• ​Waylon, Head of Platform SF Compute

• ​Marius, Member of Technical Staff, Anthropic

​About the organizers

​Tzafon

​Tzafon trains superhuman computer use action models that continuously learn.

​KERNEL

​KERNEL builds crazy fast, open source infra for agents to access the internet

​SF Compute

​SF Compute runs large-scale, vetted GPU clusters. Need H100s for an hour? B300s for two years? You got it. Need to cancel? Just sell back what you don't need.

​Nvidia

​Hardware and software for model training and inference, supports this event with compute and their open source runtime ecosystem such as OpenShell.