Aquin BountyShip
3 weeks to build,
audit, and publish
real ML work.
A fellowship for ML researchers and engineers who want to reverse-engineer LLMs, fine-tune models, benchmark checkpoints, and train SAEs, using Aquin on real projects.
why we're doing this
stress-test the platform
Aquin is a research tool. Real researchers using it for real work surface things internal QA never can. Bugs, missing primitives, workflows we never imagined.
manufacture case studies
Each fellowship project, if scoped well, doubles as a case study for one of our ICPs. Sales currency we don't have to fake.
build the community
People who use Aquin to publish a paper, train an SAE, or audit a public model become evangelists by default. We're not buying that, we're earning it.
what fellows get
- Full Aquin platform access
- Compute credits
- Discord access with founding team
- Daily office hours
- The output is yours: research notes, blog posts, lit reviews, model releases, whatever form it takes
what we get
- A case study on each fellow's work
- Bug reports, feature requests, workflow data
- Community evangelism that compounds
selection
one application form
We look for two things:
Have you actually built something? Trained a model, fine-tuned one, read an interp paper end-to-end. Pick one.
Do you know what you want to ship? Vague answers get rejected. Specific ones get in.
the 16 bounties
Each one is scoped to be doable in 3 weeks by one person. Pick one, or pitch your own.
Data Audit
Model Audit
Model Diff
Open / Build Your Own
the 3 weeks
Onboard
Discord opens. Platform access, compute credentials, bounty briefs, recorded walkthroughs. First office-hours call Sunday night.
Build
One Aquin workshop. Open calendar with the team, book calls if you need help. Friday demo: every fellow shows what's working in 5 minutes. No slides.
Ship
Friday: final demo, writeup published. Feedback before anything goes live.
Not sure if Aquin is right for you?
Aquin