Ask Claude to compare WHOOP metrics with lab results

A developer used Claude to answer questions across two years of WHOOP data and lab PDFs stored in Databricks tables.

Done with Claudeยท Intermediate

What they did
They kept two Databricks tables: one from a WHOOP export and one from blood-test PDFs that had been parsed into rows. A small Claude agent was swapped in to handle questions by writing SQL, running it through a read-only query tool, and asking follow-up queries when needed. They then compared its answers with a Databricks Genie agent on 30 questions, using a Claude Opus judge and hand-checked SQL as the reference.
How it went
Before a data correction, both agents missed every question affected by two date errors in the gold table. After fixing that issue, Sonnet improved from 47% correct to 90%, while Genie moved from 47% to 67%. The test was limited: one run per setup, 30 questions total, and Claude graded Claude.
Tips
This works best when the underlying tables are clean and dates line up correctly. Treat the outputs as analysis support, not medical advice, and verify any result that depends on time ordering or lab timing.

Try it yourself with Claude

You are helping me compare my [WHOOP / wearable] data with blood-lab results stored in tables. Use only the read-only SQL tool to inspect the data, and if needed, run more queries to check your work. Answer this question: [insert question]. Show the key numbers you used, explain any uncertainty, and say when the evidence is too thin to be confident. Before you do anything irreversible such as sending, submitting, deleting, or changing data, pause and ask me to approve the exact action. End with 3 follow-up questions I could ask my doctor.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on github.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.