Databases, consensus, ML inference and quant tooling in C++17 and Python. Every project ships with its tests, its measurements, and the exact regime where it stops working.
A working shell. Replay recorded test and benchmark runs, or type help and look around: ls, cat raft-mvcc, whoami, sudo hire-me.
Nothing reaches this page until it has been attacked, checked by an oracle that shares no code with it, and pinned to an exact commit.
Click a node to crash or restart it. Partition the leader into a minority and try to write. It can't commit — that's the invariant raft-mvcc checks with 598 assertions. A teaching simulation of the protocol, running in your browser.
Finite differences need 2n+1 revaluations. One reverse sweep returns every Greek at a constant multiple of one pricing. Values interpolated from the published 40,000-path benchmark.
Read the full case study →Two games built on the engines from my repos. Every number on screen comes from the real code, tested against the originals in CI.
An options-desk game. You inherit a book that is short volatility, with ten trading days to the close. Hedge it, work the phones, survive the news. Every Greek on screen comes off a reverse-mode tape, and one of the numbers on the screen is wrong anyway. Finding it is part of the game.
Everyone trades one live limit order book. Buy low, sell high, don't get margin-called. Four minutes, eight seats, last trader standing.
Over a year turning factory sensors, alarms and ERP records into pipelines and dashboards that engineering and operations rely on.
Looking for systems, ML infrastructure or quant roles starting 2027.