My AI Found I Was Overtraining — From My Own Data

What You'll Learn

• How to build a personal health data platform using Claude Code with zero traditional coding
• Why cross-referencing multiple health data sources (DNA, blood, wearables, workouts) produces insights no single app can
• How whole genome sequencing analysis works when you feed a VCF file to AI with access to six medical databases
• A free-text food logging approach that replaces expensive nutritional database APIs
• Why conversational AI beats dashboards for personal health analysis

Timestamps

• 00:00 Introduction
• 01:30 Mike's background — Facebook VP Eng, Deliveroo CPO, Monzo CPO, now Sling Money
• 03:02 Building a food tracker in 10 minutes with OpenAI API
• 03:59 The overtraining detection — AI correlates Oura, blood tests, and workout logs
• 05:30 Whole genome sequencing analysis via Claude Code and six databases
• 08:23 "Why build a UI when you can just have a bot analyze all the data?"
• 09:46 Claude vs ChatGPT — "jumpy first-year resident vs 40-year practitioner"
• 11:00 Layering data sources: Oura Ring, Withings (13+ years), Strong app
• 13:30 The FatSecret problem — $1,500/month for UK nutritional data access
• 15:00 MCP server architecture for health data
• 16:50 Why Mike is returning his Whoop — API limitations
• 18:00 Open Claude for WhatsApp-based health summaries
• 19:24 "One of the most exciting things to happen in the last hundred years"
• 20:30 Proactive daily health AI — the next frontier

Key Takeaways

• Cross-source correlation is the real value — No single app can spot what emerges when you combine DNA, blood work, wearable data, and workout logs in one place.
• Zero-code platforms are real — Mike built this as a side project with prompts, while running a startup and raising three children.
• Conversational AI beats dashboards — Asking questions of your data produces better insights than staring at charts.
• API openness determines hardware value — Oura's "phenomenal" API earned praise; Whoop's limited access earned a return.
• AI models have clinical temperaments — Claude tends toward cautious over-flagging; ChatGPT provides calmer, more measured analysis. Use both.
• The analysis layer is where value accrues — Health data inputs are becoming commoditised; the intelligence layer on top is what matters.

Tools & Resources Mentioned

• Claude Code — AI coding agent used to build the entire platform
• OpenAI API — Powers the free-text food tracker
• Oura Ring — Wearable for sleep, HRV, readiness (praised for API quality)
• Withings — Smart scale and blood pressure monitor (13+ years of data)
• Strong App — Workout logging
• DNAComplete.com — Whole genome sequencing provider
• SNPedia — Genetic variant database
• ElevenLabs — Voice synthesis (planned integration)
• FatSecret — Nutritional database (mentioned, replaced due to cost)
• Cal AI — Food tracking app (mentioned, replaced)
• MyFitnessPal — Food tracking app (mentioned, replaced)

Disclaimers

• Genetic analysis discussed in this episode is for informational purposes only — consult a genetic counsellor before making health decisions based on DNA data.
• AI health analysis should complement, not replace, professional medical advice.
• The Lewy body dementia segment references a serious genetic finding — viewer discretion advised.