NHScribe

Medical Efficiency App

Comment

Inspiration

One of the biggest bottlenecks in NHS patient care is the administrative backlog, with up to 2 of every 5 admin days spent typing and sending results letters manually. Meanwhile, strict patient data confidentiality laws prevent hospitals from using cloud-based LLM tools to automate this process. We wanted to build something that helps doctors return results faster while keeping all data fully private and on-site.

What it does

Our application generates patient results letters locally, directly from haematology test data. It runs entirely on a Raspberry Pi, ensuring no data ever leaves hospital premises.

Doctors can:

  • Upload or enter patient details and test results
  • Automatically generate a results letter using a locally hosted AI model
  • Edit, approve, or reject the letter before sending
  • Export it securely as a PDF

This reduces admin workload and speeds up patient communication without compromising privacy.

How we built it

How We Built It

Frontend: React webapp for clean letter review and approval interface

  • Backend: Local FastAPI server running on a Raspberry Pi
  • LLM: Lightweight on-device text generation for result interpretation and letter drafting
  • PDF Generation: Automated through Python ReportLab
  • Data Flow: CSV or JSON input → secure local processing → editable output letter

Challenges we ran into

  • Network setup when porting to raspberry pi
  • Implementing editing of PDFs
  • Integrating our work, particularly between FastAPI endpoints and React Frontend
  • Prompt Engineering Ollama Model

Accomplishments that we're proud of

  • Built a fully functional end-to-end prototype that runs 100% locally on a Raspberry Pi — with no external network calls or cloud dependencies.

  • Successfully generated, edited, and exported real NHS-style results letters directly from sample haematology data.

  • Demonstrated that AI tools can comply with strict Healthcare data governance while still saving staff time and improving patient communication.

What we learned

  • How to deploy AI responsibly in highly regulated environments, balancing innovation whilst also considering real world limitations.

What's next for NHScribe

  • Model fine-tuning on anonymized NHS letters to improve medical tone, accuracy, and consistency.
  • Expanding beyond haematology to support other test types like biochemistry and radiology reports. -Building an offline results dashboard for clinicians to track pending letters, approvals, and turnaround times.

Built With

Share this project:

Updates