I'm a self-taught programmer using Python to explore where data, machine learning, and finance meet. I learn best by building, so I set myself a simple rule: ship one real, working project every week — no filler, no tutorials copied out of a book. Just things I actually made and can explain.

It all lives in weekly-builds, a running log of that weekly habit. Each project is small but complete: clean code, a proper README, charts, and a test suite that runs offline. The thread tying them together is quantitative finance — turning raw market data into decisions I can actually reason about.

The projects deliberately build on each other:

• Stock Toolkit — where it started: pull a stock's price history and measure its return, volatility, and worst drawdown.
• Portfolio Lab — from one stock to a whole basket: correlation between assets and the mean-variance optimization behind the classic "efficient frontier," including the realistic no-shorting version.
• Honest Backtester (its own repo) — from what's optimal in theory to what actually happens: simulate holding a portfolio through real history with rebalancing and trading costs, then defend against the three ways a backtest fools you — no lookahead bias, walk-forward validation that measures the "overfitting tax" of tuning parameters, and bootstrap significance testing that asks whether the Sharpe is distinguishable from luck at all. Grew over four weeks into a small framework; pip-installable and CI-tested.

From there the work branched into machine learning:

• Star Signals (its own repo) — my first ML project: does GitHub reward real substance or surface signal? I collected 2,400 repositories, split their features into those two camps, and let a model measure which actually predicts stars. (Substance wins — the biggest predictor is simply whether a project is still being worked on.) Pip-installable and CI-tested.
• Market Structure — unsupervised learning: hand a clustering algorithm nothing but daily price moves and watch it rediscover the market's sectors — then measure, honestly, where it agrees with reality and where it finds something deeper.

Each week tends to fix or extend the one before it. I like turning a limitation I noticed into next week's project.

• Real over flashy. I'd rather ship something modest that runs and is tested than something impressive that only works in a screenshot.
• Honest results. A backtest is a hypothesis, not a promise. I try to report the costs and caveats, not just the good numbers.
• Explain it simply. If I can't describe what a project does in plain English, I don't understand it well enough yet.

Python, NumPy, pandas, SciPy, scikit-learn, and matplotlib — with pytest keeping it honest.

More original, question-driven projects across finance and machine learning — and polishing the strongest builds into standalone repos, the way Star Signals and Honest Backtester grew out of the weekly log. The habit keeps going.