Fifth year medical student at UCL. Interested in neurosurgery, ML architecture research and AGI. Back to the desk
Research
Projects
- ECG Lab
STEMI territories you can see.
- ClinicalGym
The health record as an RL environment.
- OmniMed
Where this desk came from: a study app for finals.
- Plate
Lecture notes, served on a plate.
- Snellen chart
A free, correctly sized eye chart (6/6).
- autoCCEP
Closing the CCEP feedback loop from days to about a minute.
- MEG encoder
Self-supervised contrastive learning on resting-state MEG.
- EEG in Parkinson’s
Band-power features: Parkinson’s on medication vs healthy controls.
- RepoGym
Your engineers’ Claude Code sessions, turned into a private RL gym.
- ROSE
Recursive Online Skill Evolution: continual learning for coding agents.
- Bunshin
Shared learning and context for a company’s agents.
- Endplate
One narrow channel between you and your teams of agents.
- MiMo explorer
Browse RL environments: files, office documents and the chat beside them.
- Research Taste Lab
200 exercises for practising AI research judgment.
- Surgical Worlds
Genie 3 for surgery: play simulated laparoscopic operations.
- In-Context Pruning
Usefulness masks for pruning an LLM’s context.
- trickle
Runtime types, tensor shapes and crash-time values, as you code.
- scripttype
Write type-level TypeScript as if it were JavaScript.
- snakescript
Python inside HTML, run in the browser in place of JavaScript.
- radiator
Runtime type intelligence on hover: the real type and example data.
- notmonad
Pipeline-style data transformations, chain(val)(f)(g)().
- code-haircut
Give your codebase a haircut: dead code out, by execution trace.
- Gambit
Finds the weaknesses that recur in your games, and drills them.
- chessli
Opens a finished Chess.com game as a Lichess computer review.
- claudfish
Stockfish says what is best. Claudfish lets an agent work out why.