About
I'm Dylan Siegel.
I run QuantDev.ai.
I build trading algorithms from small neural networks. I train them on tick data, test them out of sample, and publish what survives, failures included. QuantDev.ai is that work in public, plus a free course that takes you from zero to doing it yourself, plus the finished systems for sale.
The story
Why the research is a course.
“Conclusions over claims.”
Most public trading content jumps straight to claims: performance numbers, equity curves, screenshots. It rarely shows the data, the test, or the evidence behind them. I started QuantDev.ai because that gap is the work. What separates research from storytelling is whether the data was actually asked the question.
So everything here runs the other way around. Every study names the data it used, the outcome it was trying to predict, and the result it got. Ideas that fail are written up as evidence, not deleted. Pass or fail, it gets published.
I also think this material is taught badly. Textbooks assume the maths, forums assume the folklore, and almost nobody shows the whole path from an idea to a program that trades. The course is my attempt to distill what I know into that path: start with no background, finish able to research, backtest, and build your own trading algorithms with AI coding tools doing most of the typing, the way I do.
Beyond the course sits the research, which is where I work now: small neural networks that read the raw order flow. A rulebook that conditions on every state of the market ends up with a thousand settings nobody understands. A network small enough to inspect can carry those conditions and still be explained by how it behaves. Whether that produces a signal that clears real costs is the question every note tries to answer.
The systems for sale come out the far end of that pipeline, and only after they have passed every test. The course and the research are free. The systems are not.
Focus
What the work is.
Four parts: the research at the front, the course that leads into it, the open-source code that runs it, and the systems that come out the other side.
The course
Four parts, from how a market works to a running algorithm. Each part starts where the last one ended, and nothing assumes you already know the words.
The research
Small neural networks trained on real tick data: order flow, order books, trades, and price. The unit of work is a testable prediction with a stated result, net of costs.
Open source
The research harness, the studies, and the diagnostics are on GitHub. The process is visible, including the parts that did not work.
The systems
Research that passes every test becomes a closed-source trading algorithm: signal, rules, execution, and monitoring, with a live paper-trading preview.
Operating principles
How I work.
Six rules, in priority order. When two of them conflict, the lower number wins.
- Evidence before opinion.Every study names its data, its target, and its result. Failed ideas are documented as evidence, not deleted.
- Predict first, profit second.A signal has to predict something before it is allowed to trade anything. Then it has to clear real costs. No skipping stages.
- Start from zero.If a beginner cannot follow a note from the first paragraph, the note is not finished.
- Simple ingredients or nothing.If a raw measurement has no predictive power on its own, no network built on top of it will rescue it. Complexity is earned, and a model stays small enough to explain.
- Open in public. Packaged in private.The course and the research are open. The strongest surviving systems, with their tuned settings and execution logic, are sold as closed-source packages.
- The process is the edge.One strategy is not the moat. The harness, the tests, and the failure log are what make the next signal cheaper to find.
Now
What I’m doing now.
- Researching
- Small neural networks on tick data that hold a signal for seconds to minutes
- Writing
- The research notes as each run finishes, and the course part by part
- Stack
- Go research harness, small networks trained locally, AI coding agents, static TypeScript site