NHS-PATHFINDER

Inspiration

Two weeks ago, my mum needed a blood test. Her GP gave her a form for St Heliers Hospital with a 2-week wait. We found a walk-in centre in Croydon that might be faster. But no one could tell us if they would accept her form — not the NHS phone line (they hung up), not ChatGPT (it hallucinated the wrong location), not the hospital website (no information).

We drove to Croydon. We wandered looking for "pink zone" (ChatGPT said pink — it was actually blue). We found a ticket machine. We waited 15 minutes. Finally, a nurse said: "This says St Heliers. You can't do this here."

That moment — the waste, the confusion, the helplessness — is why I built NHS Pathfinder.

What it does

NHS Pathfinder is an AI-powered assistant that helps patients navigate NHS letters and find where to actually go for their tests.

• Take a photo of your NHS letter, or just type "I have a blood test at St Heliers"
• AI extracts: hospital name, test type, restrictions, deadline
• AI tells you: where to go, step-by-step directions, what to say if questioned, and a fallback phone number

No more "sorry, wrong place." No more wasted hours.

How I built it

Tools & Technologies:

• Cursor - AI code editor that wrote my entire frontend and backend. I know zero code — Cursor built everything from my plain English prompts.
• Manus AI Agent - The brain. Manus reads NHS letters, extracts structured data, and applies rules to give personalised guidance.
• Z.ai / GLM - Vision model backup for OCR text extraction from photos.
• Node.js + Express - Backend proxy to keep API keys secure.
• HTML/CSS/JS - Clean NHS-style chat interface with camera upload.

The Architecture:

Challenges I faced

1. No coding experience. I've never written JavaScript. Cursor became my pair programmer. Every bug was fixed by describing it in plain English.

2. API credits. My Z.ai key had zero balance. I pivoted to Manus as the primary engine and built mock fallbacks.

3. Security. I learned why API keys must stay in the backend (.env) — never exposed in HTML.

4. Cursor Pro limitation. Custom models require Cursor Pro. I worked around this by having my web app call APIs directly (not through Cursor's chat).

5. Deployment complexity. Getting a Node.js backend on Vercel with serverless functions took multiple iterations.

What I learned

• AI agents like Manus can read images, extract text, and reason about NHS rules in one workflow.
• Backend proxies are essential for API key security.
• Mock first, integrate later — build the UX with mock data, then connect real APIs.
• Hackathons are about problem-solving — not perfect code. A working demo with clear value beats perfect architecture.

What's next for NHS Pathfinder

• Live NHS wait times via API integration
• Support for more test types (X-ray, MRI, ultrasound)
• SMS/WhatsApp bot for patients without smartphones
• Integration with NHS App for authenticated access

Built With

• cursor