Programme Thesis
The Transfyr AI Fellowship funds a one-year, full-time, in-person research fellowship for exceptional graduate students and early-career researchers to lead a project at the intersection of frontier machine learning and real-world scientific execution. It exists to build the world's largest multimodal dataset of scientific execution and to develop ML systems that can reason over the tacit, messy, and multimodal reality of scientific work—across people, instruments, protocols, and outcomes—to make science more observable, reproducible, and automatable.
Selection Criteria
- Strong ML fundamentals (e.g., deep learning, multimodal learning, sequence modeling, uncertainty quantification)
- Evidence of exceptional research taste (e.g., pre-registration, rigorous evaluation, novel problem selection)
- Comfort with ambiguity and curiosity about how science actually happens in physical environments
- Ability to commit full-time for 12 months (current students must self-certify leave)
- Fit with the fellowship's research themes: multimodal reasoning, conflicting evidence, long-context memory, real-world evaluations, tacit expertise, physical-world AI, biosecurity/bio-risk observability, scientific execution ontology
- Potential to produce a publishable or demo-grade research artifact (benchmark, dataset, prototype)
- Willingness to work in-person in Boston/Cambridge (remote by exception)
- International applicants welcome; visa support available
- Rolling review; apply by August 15 for full consideration
Past Winners / Cohort Profiles
No specific past winners are named on the page. The programme is new (2026-2027 cohort). Based on the description, past/ideal fellows are likely graduate students or postdocs in ML, computational biology, robotics, or HCI who have demonstrated research independence and a taste for tackling messy, real-world scientific problems. They likely have strong engineering skills and a portfolio of projects that show both technical depth and scientific curiosity.
Ideal Candidate Fingerprint
The platonic ideal applicant is a current PhD student or postdoc in ML/computational biology with a proven track record of rigorous, pre-registered research that bridges ML and scientific practice. They are comfortable with ambiguity, have strong software engineering skills, and are excited to work with large multimodal datasets of scientific execution. They can articulate a focused research question that aligns with the fellowship's themes and demonstrate the ability to build prototypes and evaluations that have real scientific impact.
Recommended Framing
The strongest angle for Eniola is to leverage the TOPOLOGIX project as the core research proposal, because it directly matches the fellowship's focus on building ML systems that reason over scientific execution data (here, protein sequences and drug-resistance outcomes) and on real-world evaluations (benchmarking against Platinum and SKEMPI). Eniola should frame TOPOLOGIX as a foundation for a larger vision: using multimodal scientific-execution data (e.g., lab protocols, instrument traces, outcomes) to predict and understand drug resistance in real time, aligning with the fellowship's 'scientific execution ontology' and 'real-world evaluations' themes. The rigorous pre-registration and honest reporting of negative results (e.g., hERG topology study, ergofluids) demonstrate the 'exceptional research taste' the fellowship seeks, and the multi-domain expertise (pharmacology, ML, dynamical systems) is a unique asset for tackling the ambiguity of real scientific workflows.
Watch Out
- Full-time in-person requirement in Boston/Cambridge may conflict with M.Sc. enrollment at HPI/Potsdam (though remote by exception may be possible, but not guaranteed).
- IP ownership clause: Transfyr owns fellowship work product, which may be a concern for an independent researcher with multiple ongoing projects and potential future ventures.
- Applicant is not a current graduate student in ML/CS/computational biology (though enrolled in M.Sc. Digital Health, which is adjacent), and the fellowship targets 'current graduate students able to take leave, recent PhDs, and postdocs'—Eniola is pre-PhD, which may be a competitive disadvantage.
- The fellowship is highly competitive and focused on frontier ML; Eniola's background is more applied and multi-domain, which may be seen as less 'frontier' unless framed carefully.
Research History
2026-08-04 20:11 · medium confidence
2026-07-31 17:18 · medium confidence