Pick a learning track

A model call is stateless, so every useful system is a loop you own. Build the loop, then everything hard about it: tools, memory, retrieval, orchestration, and how you know any of it worked.

A pipeline is a promise about data that has to keep being true tomorrow. Read a messy file without a library, quarantine what is wrong instead of dropping it, put it in a table you can query, then make the whole thing re-runnable and checkable.

A backtest is a hypothesis about market mechanics that has to survive execution friction. Build a limit order book, quantitative time series signals, an event-driven backtester, options pricing models, portfolio risk systems, and statistical validation that resists overfitting.

Programs translate algorithmic intent into memory operations across distinct runtime architectures. Master low-level bit representation, memory layouts, and manual allocation in C; compile-time type safety, RAII, and zero-cost polymorphic abstractions in C++; and dynamic object graphs, bytecode interpretation, and metaprogramming in Python.