Home / Fellowships
Fellowships
Three to twelve months of funded time to commit deeper to one well-scoped research thread — the kind of work where the bottleneck is time to think, not access to a bigger cluster. Stipend, compute, mentor pairing, and a cohort of other independents at the same career stage. Fully remote. You keep your IP and publish under your own name.
- Stipend
- $30,000 / year (pro-rated)
- Compute
- $25,000 cluster + cloud credit
- Duration
- 3 – 12 months
- Cohort
- 8 fellows / year
What it is
A fellowship is for independent researchers who want time, money, and compute to dig into one well-scoped research thread. You stay where you are. You keep your IP. You publish under your own name.
It is not a PhD substitute, an internship, or a job. It's a focused stretch of work, with the friction of compute, rent, and conference travel removed.
What you get
- $30,000 stipend annualised, paid monthly
- $25,000 in compute credit on the alphabell cluster + cloud passthrough
- One mentor drawn from L8+ active members, paired by topic, meets monthly
- Cohort access — weekly seminars, dedicated Discord, retreats at months 4 and 10
- Travel budget up to $3,500 / year for conferences, retreats, or visiting members
- Editorial support for writing up your work (we ship copy-editors, not co-authors)
What we ask
- Publish at least one piece of work in the year — paper, dataset, tool, audit, or substantial open-source release
- Co-host one alphabell event during the year (a competition, hackathon, or reading group)
- One short monthly update on the cohort Discord
- Open-source by default; if your output is best closed, we want to hear why up front
Who's a fit
Independent researchers, data scientists between jobs, late-stage PhD candidates, hackers with a research project they can't get to in their day job, and people in countries with no AI research scene of their own. Working scientists at frontier labs are not eligible — they have the resources already.
Applications close 15 Aug 2026
Eight fellowships available. The application is a short PDF plus three short questions. No CV.
3-month residencies · rolling
For members who want a shorter, project-scoped block of fellowship-level support.
2025–2026 cohort
Current fellows
Mira Holloway
Bay Area · US
Durable agent substrates: persistent state, learned tool affordances, verifiable execution traces.
Sofia Karvelas
Athens · GR
Mechanistic study of refusal directions in open-weight small models — when and why they generalise.
Cheung Wai-Lin
Hong Kong
Adversarial-prompt corpus for Cantonese tool-use — the first public dataset of its kind, 18k labelled examples.
Dineth Karunaratne
Colombo · LK
Replication and audit toolkit. Reproduced nine published router-distillation papers end-to-end this year.
Henrik Lindqvist
Göteborg · SE
Causal scrubbing for tracking feature lifetimes across pretraining checkpoints.
Adaeze Okeke
Lagos · NG
African-language code-switching prompt set — first of its kind for agent tool-use.
Tomasz Włodarczyk
Kraków · PL
Reproducible KV-cache compression benchmark across six long-context open models.
Léa Bouchard
Montréal · CA
Attention-sink dynamics in long-context models — code and checkpoints released monthly.