I build agent systems and the tools to understand whether they actually work.
Currently an AI Specialist at Solace, working with event-driven AI. Previously at SAP and Deeplink Medical. Based in Paris.
| Project | What to look at | Status |
|---|---|---|
| Backtests as Reward Signals | Controlled evaluation of false discoveries in autonomous financial research; frozen evidence and reproduction checks | Working-paper artifact |
| Lodestone | A Rust execution log with byte-exact replay and recorded model/tool effects | Replay-core prototype |
| AI Adoption Lab | Bounded workflows, human approvals, retrieval, evaluations and traces | Local demonstration lab |
| Numerai Solstice | Quantitative ML: tree and transformer ensembles, era-level evaluation and neutralization | Research implementation |
| Solace Trading Lab | Event-driven simulation, order books, fault injection and replay | Infrastructure simulator |
| Origamify | A browser-based crease-pattern editor, where programming meets paper | Interactive prototype |
I like understanding the failure modes as much as building the happy path. Repository documentation separates implemented behavior, simulations, research results and future plans.
TensorTonic is my collection of synchronized ML exercises. Almanac explores deterministic software in a small e-ink object. Away from the keyboard: origami and running.
I enjoy collaborating with people who care about useful software, careful experiments, and making complicated systems understandable.