What I am exploring
Exploring long-term questions across biological systems, computational models and intelligent systems.
Active / Building
How do you make a group of agents reliable enough that a person stops reading every intermediate step? The question is about trust in the run as a whole, not about making any single agent cleverer.
Related projects
Can physiological state be estimated from noisy, sparse, everyday signals — and described without overclaiming? This sits where health products meet biological modelling, and its hardest constraint is knowing where to stop.
When is a simulated trajectory useful, and when is it an expensive-looking guess? The interesting part is not producing the trajectory but being able to say which of the two you are holding.
Related projects
What does personal software look like when the user owns the data and the graph never leaves their device? The direction treats the device — not a server — as the place where personal intelligence lives.
Related projects
Concepts / Exploring
What would an AI system have to do before a scientist would spend real money testing the hypothesis it proposes? The direction is about that bar — proposing something worth testing, and keeping what was observed separate from what was assumed.
Related projects
Lab Notes
Short notes on what changed in the public record. No roadmap, no internal design.
2026.09
Snapshot
The open MVP still reads the same: four clocks, the organ-clock layer, the integrator and the agent layer are all in the tree. The causal graph remains marked as partially open.
2026.09
Note
The GSE130973 subset (5,000 cells × 2,000 genes) has an inspect → prepare → validate → summarise → smoke-train path with its own report. Age and state labels stay explicitly unknown.
2026.09
Release
Three themes, a Chinese-first entry, and a mathematical-biology section built from original diagrams. The fact layer underneath did not move.