Domain Fellows
Practitioners who use Raycaster on real or de-identified work, review outputs the way a peer would, and help define submission-grade acceptance conditions.
AI Fellows program →Experts · Fellows · Partners
Experts don’t label chat. They make work measurable.
Raycaster’s evaluation records are only as good as the domain judgment inside them. Fellows, practitioners, and specialist partners turn consequential workflows into tasks, rubrics, and verdicts—while protecting confidential material.
Looking for paid work? We fund researchers for two to three months to build a benchmark in their own domain, and they keep the credit for it.
Sponsored benchmark research →How Raycaster works for you
Your expertise becomes infrastructure.
This is not a generic annotation marketplace. Contribution means capturing real professional work, specifying what “good” looks like, and adjudicating whether an agent actually completed it.
Roles
Choose how you participate.
Campus Fellows
Students in regulated and technical programs who learn AI-native professional practice while contributing structured review and example packets.
Campus Fellows program →Expert partners & firms
Specialist shops and domain networks that supply representative workflows, adjudication, and QC at program scale—with clear confidentiality boundaries.
Partner intake →Adjudicators & rubric authors
People who translate tacit standards into observable criteria, resolve grader disagreements, and keep holdouts honest.
Contribute judgment →The loop
From tacit skill to inspectable verdict.
01
You bring the work
Representative tasks, source packets, edge cases, and the decisions a competent practitioner would actually make.
02
We instrument the environment
Files, tools, and state become a reproducible workspace—usable for evaluation, training episodes, and regression.
03
Judgment becomes the dataset
Rubrics, graders, and expert review turn tacit standards into evidence another team can inspect and defend.
04
Public, qualified, or NDA
Selected traces can educate the market. Sensitive packets, recipes, and partner identities stay gated by design.
Confidentiality
Protect the work. Still make progress.
01
De-identified by default
Fellows programs use placeholders and sanitized examples—not your employer’s live dossier.
02
Contribution ≠ public disclosure
Partner identities, raw packets, and holdouts stay qualified or NDA unless you explicitly approve.
03
Dual benefit
Participation can improve public benchmarks and your own private evaluation / deployment readiness.
Contribute or evaluate
Supply judgment for the ecosystem—or build a private eval for your team.
Start
Tell us the domain you know cold.
After intake you’ll get a contribution guide: what packets look like, how review works, and which track—Fellows, campus, partner, or adjudication—fits.