Self-Driving Portfolios: AI Agents for Investment Research. Everything for the week, in one place.
N to unlock.Mystery dataset, backtest by hand, then the principles (lookahead, survivorship, overfitting).
The ML ladder: CAPM to trees, PCA, neural nets; LLMs and agents at altitude; week setup.
Colab template, agent-written factor models on structured data, verify what the agent wrote.
Typed JSON scorer on text, next-session signal, golden-set v1 self-run, local downloads start.
Harness anatomy, loops, memory and RAG mechanics, orchestration, the router as live example.
Evals on their own outputs, pass@k vs pass^k, reproducibility, guardrails, red-team swap.
Local LLMs in Colab and on laptops, quantization, token economics, routing and cascades.
Fine-tune vs prompt vs RAG, synthetic data, cutoff contamination + anchor reveal, bake-off, freeze.
Alpha vs operational beta, ablation and attribution, REAL-ALPHA verdicts, pitch runbook.
Holdout release, frozen-notebook live runs, team presentations, awards and close.