Patient-education assistants for sarcoma and brain health
Two AI assistants, one for sarcoma patients and carers and one for brain health, that answer from clinician-approved content with citations and route urgent questions to emergency help.
- Industry
- Healthcare and life sciences
- Services
- RAG assistant, voice input, clinician dashboard
- Stack
- Python, Flask, Anthropic Claude, OpenAI Whisper, React
- Purpose
- Patient education, not diagnosis
The problem
People facing a serious diagnosis search online and often find contradictory, alarming or wrong information. A general-purpose chatbot is risky here: it can sound confident while being inaccurate, and it does not recognise when a question is really an emergency.
How we approached it
Both assistants retrieve answers from a curated library of clinician-approved sources and return them with citations. A confidence threshold stops the model from improvising when the sources do not cover a question. An emergency layer recognises urgent language and points people to real help, and a clinician dashboard lets qualified staff review conversations and keep the content current.
What we built
Source-grounded answers
Retrieval-augmented generation matches each question to curated, human-reviewed content before the model responds.
Citations on every answer
Answers link to approved sources, colour-coded by type, so readers can see where the information came from.
Emergency detection
Urgent symptoms are recognised and routed to local emergency numbers and care information instead of a chat reply.
Voice input
Patients can speak their questions. Speech is transcribed in the chat, with fallbacks when no mic is available.
Clinician dashboard
Usage analytics, top questions and conversation review with role-based access.
Cached retrieval
Embeddings are built ahead of time and loaded at startup for sub-second, low-cost semantic search.
Tech stack
- Python
- Flask
- FastAPI
- Anthropic Claude
- OpenAI Whisper
- sentence-transformers
- scikit-learn
- MongoDB
- React
- TypeScript
- Node.js
- PostgreSQL
Results
Patients and carers get clear, sourced information by typing or speaking, and urgent situations are flagged instead of answered. Clinicians can review what the assistants say and keep the content accurate.
These assistants are designed for patient education. They do not provide medical diagnosis or replace professional care. In an emergency, contact local emergency services.
What is the Sarcoma & Brain Health AI project?
It is a pair of healthcare AI assistants for patient education, one focused on sarcoma patients and carers and one on brain health. Both use retrieval-augmented generation over curated, clinician-approved content so answers come with citations rather than guesses.
How does the assistant keep medical answers safe and accurate?
Answers are retrieved from a library of approved sources and returned with citations, a confidence threshold prevents the model from improvising, and an emergency layer recognizes urgent language and points people to real help. A clinician dashboard lets qualified staff review conversations and refine content.
Does it replace a doctor?
No. These assistants are designed for patient education and do not provide diagnosis or replace professional medical care. In an emergency, users are directed to contact local emergency services.
Building an AI tool for patients or clinicians?
Tell us about the audience and the content you trust. We will explain how we would keep answers grounded and safe.