Most "AI for investing" demos quietly backtest on historical data and never risk a dollar. This one does the opposite: it hands ChatGPT $100 of real money and lets it run a live micro-cap portfolio for six months, logging every decision so the track record can't be cherry-picked after the fact. The question it actually probes isn't "can an LLM beat the market" but "what does an LLM do when forced into repeated, accountable financial decisions under hard constraints."
What Sets It Apart
- Real money, real time: trades run on live micro-cap equities with a strict rule set and automated stop-losses, so survivorship and hindsight bias have nowhere to hide.
- Full decision transparency: each trade ships with the model's written rationale, plus daily CSV portfolio snapshots and weekly deep-research artifacts — you can audit why it bought, not just what it bought.
- Quant-grade scoring: results are graded with Sharpe, Sortino, CAPM and drawdown metrics and benchmarked against the S&P 500 and Russell 2000, not vague "it went up" claims.
- Reusable engine: what started as one experiment hardened into a Python framework (pandas, yfinance, Matplotlib) others can fork to run their own model-vs-market trials.
Who It's For
Great fit if you want an honest, fully documented case study of an LLM acting as a portfolio manager, or a starting harness for your own live or paper-trading runs. Look elsewhere if you need a deployable trading bot or financial advice — this is a transparent research log of a single $100 wager, and the micro-cap volatility plus one-account sample size make the results illustrative, not statistically conclusive.
