My partner (ex-Microsoft) and I were responsible for every aspect of a hedge fund management business:- team organization- goal setting- hiring- individual code contributions- code reviews- research reviews- project specs- trading operationsWe broke down the hard problem of managing an equity market-neutral hedge fund into several layers:- extract (data engineering to power research + trading operations)- predict (a family of ML models predicting individual stock alpha)- solve (portfolio optimization)- trade / validate (real + simulated trading environments to evaluate portfolio performance and calibrate simulator assumptions)Tech stack / skills involved:- lots of python (pandas, numpy, jupyter notebooks etc)- docker images at each layer to capture reproducible behavior from research in our isolated production trading environment- terraform to make sure AWS environment was reproducible- portfolio optimization (cvxpy)- workflows / CI / CD (AWS step functions, github actions)- ELT / ETL (Snowflake, DBT, AWS Athena, python scripts)ML for Financial Timeseries Forecasting:- Used walk-forward prediction and several stages of out-of-sample validation (including leak testing) to evaluate model performance, overfitting - underlying models ended up being mostly boosting algos (catboost, lightgbm)- Compute (AWS Sagemaker)