What it does

This repo documents a live experiment: ChatGPT managed a $100 micro-cap stock portfolio for six months under hard rules, with every trade, daily update, and weekly research chat preserved in CSVs, PDFs, and Markdown. The code is a Python trading engine (yfinance + pandas + Matplotlib) that enforces stop-losses, logs decisions, and benchmarks against the S&P 500 and Russell 2000. A 40-page evaluation paper is included.

The interesting bit

The author started this because an AI stock-picking ad annoyed him. Rather than rant, he made the LLM put its money where its mouth was — literally. The repo’s real value isn’t the trading engine; it’s the obsessive transparency. Historical artifacts are frozen; new experiments layer on top without rewriting the past, turning a stunt into a reproducible research baseline.

Key highlights

• Forward-only experiment record: no back-dated heroics, all logs public
• Production trading script with automated stop-loss enforcement
• Full decision audit trail: weekly deep-research chats, trade logs, portfolio snapshots
• Risk analytics included: CAPM, Sharpe, Sortino, drawdown metrics
• Reusable framework structure; author spun off LIBB for general LLM investor-behavior research
• Next experiment: IPO-focused, with monthly Substack analysis

Caveats

• The $100 starting capital and micro-cap focus mean results may not scale; the README doesn’t discuss slippage or liquidity constraints
• Tech stack is deliberately plain (yfinance as primary data source, CSV accounting), which is either refreshing or a limitation depending on your needs
• The 40-page paper’s conclusions aren’t summarized in the README; you’ll need to open the PDF

Verdict

Worth a look if you’re studying LLM decision-making under uncertainty or building reproducible trading experiments. Skip it if you want a production quant framework — this is a lab notebook with code attached, not a trading platform.