Healthcare systems are notoriously bad at sharing information between themselves. Your GP has your prescriptions, the hospital has your discharge notes, that specialist you saw three years ago has their own siloed record—and good luck getting them to talk to each other. A new write-up on OpenHealthHub demonstrates a different approach: build your own comprehensive health record using an LLM as your organizer and analyst, keeping everything local under your control.

The Setup

The experiment requires VS Code and the Claude Code extension (search for it in the Extensions panel or use the installer). You'll need a Claude subscription—Pro at roughly £20/month works fine. Before anything else: go to Settings > Privacy and disable 'Allow the use of your chats and coding sessions to train and improve Anthropic AI models.' That opt-out is on by default, which is exactly backwards when you're feeding it sensitive medical data. Create one folder for everything—say Documents/health—and open it in VS Code. Claude can now read and write inside that folder, and only that folder.

Collecting Your Medical Data

The workflow starts with aggregation: drag in PDFs of hospital letters, discharge summaries, clinic notes, blood results, scan reports, optician prescriptions, vaccination records, occupational health certifications—anything and everything. Screenshots of paper documents work too. For records you don't have yet, the author recommends drafting Subject Access Requests under UK GDPR Article 15 to every GP practice, trust, and clinic you've used. Ask for imaging studies themselves, not just the reports. The Claude Browser Extension can automate extraction from painful web UIs like the NHS App.

Organizing With LLM Assistance

Once everything is in one place, ask Claude to transcribe each PDF into structured Markdown files while keeping originals untouched. A first prompt like 'Read every file in this folder, transcribe each PDF letter into a structured Markdown file next to the original, don't delete anything, then build me a current problem list, medication list, and chronological blood-results table' gets you started. You decide the structure—purely chronological, problem-oriented, or hybrid. The key insight: 'LLM hallucination is more likely to occur when there is an absence of factual information, and LLMs have to make up the gaps with bluff in order to please you.' Feed it comprehensive data, and accuracy improves dramatically.

Mining For Insights

Now your record stops being a filing cabinet and becomes a tool. Ask questions across your entire history: 'Plot my blood pressure trend,' 'What immunity is actually documented versus inferred?' 'Which results have drifted over ten years?' 'What's missing that I should chase?' The author notes the LLM seems particularly capable with rare diseases—its broad training data lets it point out possibilities that experienced clinicians haven't connected. Individual doctors simply can't know about every rare genetic cause of a given symptom. LLMs just do, by default.

Practical Caveats

Keep originals—the LLM writes structured copies alongside source PDFs, never instead of them. Every derived fact stays traceable to its source document. Be skeptical of complex clinical calculators; simple ones like BMI are probably fine, but QRISK2-3 calculations shouldn't be trusted blindly from an LLM. Web-based calculators accessible via the Browser Extension might fill that gap. The author is apparently working on a suite of fast, LLM-friendly command-line clinical calculators to address this gap.

Key Takeaways

  • Everything stays local; you control your data completely
  • Disable model training in your Claude privacy settings before uploading anything medical
  • SAR requests under GDPR Article 15 can get you records from NHS and private systems
  • Comprehensive factual context reduces hallucination risk significantly
  • LLMs excel at spotting rare disease patterns that individual clinicians might miss

The Bottom Line

This isn't just a clever hack—it's a glimpse at what patient empowerment actually looks like when you stop waiting for healthcare systems to interoperate. Your GP probably doesn't have as complete a record of you as you'd build in an afternoon, and they certainly don't have it all searchable at their fingertips. For anyone managing chronic conditions or navigating complex medical histories, this approach turns the information asymmetry on its head.