CoreWeave Hacks: Agent Loops Hackathon with Weights & Biases and AGI House
About the event
Come plug in and supercharge at our premiere CoreWeave hackathon! If you’ve had an idea for an autonomous loop that’s been burning away at you, this heavily-resourced hack is for you to make it happen. Experience everything you’ve come to love about Weights & Biases WeaveHacks– this time with the full CoreWeave stack (Weave, ARIA, marimo, and much more).
Join us at our SF office (this is an in-person event) to have the most fun building loops you'll have this year. This round is all about the loop: building agents that cycle through reasoning and action, catching their own mistakes, and pushing towards more powerful results at each pass.
Agent capabilities have advanced enormously over the last year — even by their own doing: better memory systems, new forms of RL and fine-tuning, dynamic tool creation, and self-improving loops. This time we observe, evaluate, track, document, and self-improve agent iterations while we play with a robot dog and humanoid!
Hackers get Fully Connected
Attendees have the chance to attend our Fully Connected conference September 29-October 1 (each ticket ~$1k in value), where a final CoreWeave Hacks award will be presented for the “Most Production-Ready” submission.
Criteria for any non-winning attendee to get a ticket will be announced at the event. Once building starts, find out more from a CoreWeave representative.
The Best Loop Design winners on 9/13 get tickets and stage time at Fully Connected.
The winning team gets a stage demo on the final day, presenting their project in front of AI leadership. Build like it matters; the work can carry straight from the hackathon floor onto that stage!
You will have a second chance to win 2 weeks after CoreWeave Hacks, at our Fully Connected conference!
We bring
A killer office with ultrawide monitors and standing desks
Great food & drinks (breakfast, lunch, and dinner)
Expert judges
A humanoid that hands out stickers and dancing robot dog
The full CoreWeave SaaS stack for agents
A chance to attend Fully Connected — and a stage demo for the winners
A media team capturing the event (winner video, coverage, and interviews)
Whimsy
You bring
Your ideas
A readiness to actually show up & crank out code
Good vibes
Will you build a dojo where agents train themselves? A hall of mirrors where agents grade their own reflections? A memory palace agents keep rewriting? A forge where agents hammer out their own tools? A loop that learns from its own scars? A colony of agents keeping each other honest? However you've been dreaming about closing the loop on self-improving agents — no more excuses. This is your sign to build.
If you don't know us: CoreWeave Hacks is our AI agent hackathon series (formerly "WeaveHacks"). Our first ones drew over 200 builders from across the country to network, build, and deploy agents. Due to popular demand, the series continues with new features.
What to expect
$20k+ in prizes | Robot dog for Best Loop Design | F1 tickets for Most Production-Ready
Robot dog and humanoid
Tons of food, snacks, drinks
A chance to attend Fully Connected; a stage demo there for the winners
Weights & Biases onsite, plus additional sponsors with hands-on help integrating their tools — only a few, no hours-long presentations. We're here to hack.
Expert judges who offer specific feedback
A media team documenting the event
Schedule
Day 1 (Saturday)
9:00 AM — Doors open & breakfast
10:30 AM — Kickoff & initial presentation
11:15 AM — Hacking starts
6:30 PM — Dinner
9:00 PM — Office closes
Day 2 (Sunday)
9:00 AM — Doors open, snacks
1:00 PM — Submissions due
1:00 PM — Lunch
1:30 PM — Judging begins
3:30 PM — Project presentations
4:30 PM — Awards ceremony
Sponsors
Your host: CoreWeave
Weights & Biases — the full agent stack: Models, Weave, Sandboxes, ARIA.
Marimo notebooks — modern Python notebooks with free cloud GPUs in molab
Strategic Sponsor: AGI House — an applied AI lab, venture fund, and curated community of the founders, researchers, and engineers defining frontier AI. AGI House partners with CoreWeave and Weights & Biases to bring the best builders and the best tools into the same room.
TypeSafe AI | TypeSafe is a new AI lab that has built a new class of models designed for machine-native intelligence.
Judges
Emmanuel Turlay | Weave team, Weights & Biases | Emmanuel is Directory of Engineering at CoreWeave, focusing on agent observability with the Weave team. Previously a YC founder, and leading the ML platform team at Cruise
Lorenzo Porras | Weights & Biases
Julia Rose | ARIA team, Weights & Biases | Julia is a Staff Product Manager at CoreWeave, focusing on Models Workspaces and our in-app coding agent ARIA. Previously an ML engineer in self-driving.
Konstantin Taletskiy | marimo | Developer Relations Engineer at marimo, former Anaconda open-source engineer, and notebook ecosystem contributor with a PhD in Chemical Engineering.
Ryan Biddy | Weights & Biases
Diogo Almeida | TypeSafe AI | Diogo co-invented RLHF and InstructGPT, the methods that lead to ChatGPT and GPT4. Previously, he was at Google Brain. His research has helped shape modern language models, and he continues to push the frontier of AI systems.
Mo Tiwari | Google DeepMind | @mo_tiwari AI Researcher at GDM, previously OpenAI and Stanford CS PhD. Currently working on personalization for Gemini and AI for math.
Nirav Patel | Okta
Shivank Joshi | Mindis.ai, The Foundery | I most recently managed a family office of an Indian billioniare and was the first person on the founding team of The Foundery ( an venture studio by three billionaires), prior to that I worked on biomimicry-inspired ornithopter drones, and on weekends I host a podcast on investing and institutional building ( 199+ episodes, 21 billionaire guests, and over 120K subscribers).
Kshitij Dixit | Perit.ai (YC W20)
AGI House
Additional judges announced soon!
Named prizes
Best Loop Design — Robot Dog ($4k value) and $2k cash (all cash prizes distributed evenly between team members), presentation at Fully Connected conference.
Most Production-Ready — F1 tickets (~$1-2k/ticket) + $1k cash
Best Use of Weave — $1,000
Best Use of ARIA — $1,000
Best Use of marimo — $500
Best Social Media demo – $1,000
Best Use of TypeSafe AI — to be confirmed.
Applying to participate
Apply on Luma. We prioritize hackers who have some kind of project they've built and can point to (it doesn't need to be polished at all) — GitHub or Devpost links are fine. No specific experience required.
Given the WeaveHacks 4 demand, we will be selective with a cap at 200. We'll likely hit max capacity before the event, so apply early, and mark yourself going / not going as soon as you know so you're not holding a slot. Please do what you can to have your entire team (all members) sign up ASAP, as we likely will not have space for new team members to join at the door.
What’s holding you back? This is your sign to build.