Jorge
Projects

Trading system · 2026

Market making engine

A quoting system for thin prediction markets, built to study the part nobody writes about: what happens when the connection drops and the spread blows out.

Personal project
PythonCLOB APIRisk sizingFailure handling

What it is

A system that posts resting quotes on both sides of a thin order book, skews them as inventory builds, and hard caps exposure. Roughly a fifth of the code is the quoting logic. The rest is everything that has to hold when the venue misbehaves.

Why I built it

Market making is the cleanest sandbox I know for a problem I care about in every system I build, which is what happens at the edges. The strategy is a page of maths. The engineering is reconnect logic, idempotent order state, stale quote cancellation, and knowing when to stop trading entirely because the book has gone weird. That distribution of effort, ten percent idea and ninety percent guard rails, is the same in a payments integration and in an onboarding pipeline.

How it works

Quote both sides around a fair value estimate. As fills accumulate on one side, skew the quotes so the book naturally pushes the position back toward flat rather than relying on a separate unwind. Cap exposure so no single move can matter. Kill switch on stale data, on disconnect, and on a spread wider than a threshold, because the moment you most want to exit is the moment the venue is least able to let you.

What I learned

Backtests do not contain the failure modes. Every real lesson came from a socket dropping at the wrong moment, or from an order I believed was cancelled and was not. The fix is never cleverer logic, it is treating every external system as unreliable by default and making the code prove state rather than assume it.

Where it stands

A personal technical project, not a product and not an offer of anything. The next piece is a pre trade filter that reads a market's historical book depth and refuses to quote the ones too thin to be worth the risk.