From zero knowledge: a dev's attempt at systematic trading, in public
I’m a fullstack software engineer. C# on the backend, React on the front. I have roughly zero knowledge of finance beyond an index fund and a vague sense that “the Greeks” aren’t just a civilization.
I’m going to learn quantitative trading from scratch, build all the infrastructure myself, trade a small account systematically, and publish everything — the code, the backtests, the P&L, and especially the failures. This post is the commitment device.
Why this might not be stupid
Retail trading content is mostly noise: signal sellers, screenshot bros, courses about courses. What’s rare is someone showing working code and real numbers over a long period. As an engineer, the infrastructure — backtesting engines, data pipelines, execution systems, dashboards — is the part I’m actually equipped for. Most retail traders can’t build it. Most quants who can build it don’t write about it for normal developers.
So that’s the bet: the intersection of “can code” and “will show everything” is underpopulated, and the tooling I build along the way is useful even if my strategies turn out to be expensive random number generators.
Why this might be stupid
Let me steelman the skeptic, because the skeptic is usually right:
Markets are brutally efficient at the timescales retail can access. Fees and slippage quietly execute most backtest darlings. Overfitting is the default outcome of strategy research, not the exception — give me enough parameters and I’ll fit you a beautiful curve on anything. And a sub-$5k account means even a good strategy produces returns that round to lunch money.
All true. Which is why the honest goal for year one isn’t trading income. It’s three things: real knowledge, real infrastructure, and a real public record. If an edge shows up, great. If not, I’ll have proven that too — with receipts — and that’s worth more than most trading content on the internet.
The rules
These are fixed. Written down now so future-me can’t quietly bend them:
- Paper trading only for the first 60 days. Live money starts tiny ($1–2k) and only after the process holds.
- No day trading. Swing timeframes, days to weeks. (The PDT rule makes this decision for small accounts anyway.)
- Options are defined-risk only. Spreads where max loss is known at entry. No naked short options, ever.
- A backtest doesn’t count unless it models fees and slippage and survives out-of-sample data.
- Everything gets published. Losing trades are content, not secrets.
- None of this is advice. It’s a lab notebook.
The stack
QuantConnect’s LEAN engine — open-source, and written in C#, which for me is home turf — for backtesting and eventually execution. A data pipeline off free-tier APIs to start. This site: Astro on Cloudflare Pages. Everything lives in one monorepo, and the reusable parts get open-sourced.
I’m also using AI heavily and openly: as a tutor for the finance curriculum, a pair programmer on the engine, and an editor on these posts. Part of the experiment is figuring out how much leverage that actually buys a solo operator. I’ll report on that too.
The 90-day plan
Days 1–30: foundations — market mechanics, options fundamentals, risk math, LEAN running with a first toy backtest. Days 31–60: two or three candidate strategies coded, backtested honestly, and running on paper. Days 61–90: the best paper performer goes live at tiny size, and the success metric is process adherence, not P&L.
The full interactive version lives at /roadmap, and it’ll track actual progress against the plan.
What “success” means
Not a Lambo. Success at 90 days: I can explain a credit spread’s failure modes without notes, my own pipeline produces backtests I can defend, four weeks of live trading happened exactly per the rules, and this journal published every week without fail.
The consistent-publishing part is, honestly, the hardest system to build. That’s the real experiment.
See you in the next post — first up: getting LEAN running and losing my innocence about backtest results.
— zk