Stop Writing the Analysis Script: Let the Data Decide What to Look At
summary
A manual trading indicator built from ~75–79 million rows and 1,180 trading days becomes a fast experiment when fed to Claude rather than hand‑coding rules; within ~10 minutes the model tests ~9,500 trades, surfaces a ~48% direct predictor and a nowcast at ~77%, and moves a simulated result from −$139k to +$37k (Δ ~$177k). Uses Opus/Fable variants, requires prompts with cautions and checks (forward‑return timing bug), and warns that AI will produce plausible but sometimes misleading answers.
AI-written summary · about
chapters (6)
- 0:00Data analysis challenges
- 2:34Manual trading indicator
- 7:34Manual to AI analysis
- 11:49AI data mining demo
- 20:39AI analysis lessons
- 26:02Final insights
related talks
19mStock data analysis with
10mAnalyzing 10,000 Sales Calls With AI In 2 Weeks
28mGenerate Insights, Not Excuses: How your data is slowing you down
19mTeaching Coding Agents to do Spreadsheets
02h 04mShip Real Agents: Hands-On Evals for Agentic Applications
21mAI-Powered No-Code for Enterprise Software
18mLessons from Analysing all of Europe's Data
15mUsing AI effectively in large Java codebases