Trading architecture

Personal project · Python, Docker, GitHub Actions · Not a professional desk, and not a claim about returns

Why it exists

I solve problems with software. A few years ago the problem was: I wanted to understand market systems by building the plumbing, not by reading about it.

The interesting part was never a single indicator. It was how you run strategy code as a real system: isolated processes, a broker connection that can fail, a calendar that is not Monday through Friday in the abstract, logs you can trust, and a deploy path that does not require sitting at a terminal at 8:55 a.m.

What I built

A small architecture of loosely coupled services:

A broker gateway container as the shared connection point
Strategy services as separate Docker containers, one process per symbol, so one bad actor could not take down the rest
Scheduled execution with market-calendar awareness, not a naive cron string
Retry and backoff around a flaky broker API
A health-check service for connectivity and market-data sanity
GitHub Actions to rebuild and deploy instead of a manual ritual

The strategy code was Python. The packaging was Docker Compose. Two strategy implementations lived in that shape, plus a health-check utility. I treated it as production-shaped work even though it was only my own capital and curiosity on the line.

What I was actually practicing

Operational questions. What happens when the gateway is down? How do you avoid client-id collisions across containers? How do you keep jobs from firing on a holiday? Where does the log go so you can reconstruct a morning?

Those are the same muscles as any system that has to be correct, on time, and still standing after a bad night: services, data, deploy, and on-call thinking, just on a smaller stage.

What this is not

This was an independent architecture project. It is not a fund, not a trading desk, and not a performance track record. I do not publish returns. I am not claiming professional markets experience from it.

It is evidence of how I work when nobody assigns the problem: I design the system, I run it, and I keep going until the operations are honest.