• GIF

Aida - AI-Assisted Diagnosis for Time-Pressed Doctors

Inspiration

Appointment times in the NHS are extremely short, usually 10-15 minutes per patient. This leaves little time for doctors to fully understand symptoms, explore diagnoses, and explain care plans. Patients often leave appointments still needing answers. We wanted to create an AI-powered solution to augment doctors' decision-making and patient education within the tight time constraints.

What it does

Aida is a web app that helps doctors quickly arrive at the most likely diagnosis based on the patient's symptoms. It searches anonymized medical records to find similar past cases and provides a ranked list of probable diagnoses. Doctors can quickly cross-check their assessment. It also auto-generates a plain language care plan for patients explaining the issue, next steps, and resources for more information.

How we built it

We used Claude AI to create a natural language interface for doctors to describe symptoms. This input is searched against a database of de-identified patient records indexed using Weaviate semantic search, and filtered by the patient's medical history overlap with other patients (leveraging a secondary Weaviate collection of Claude-summarized histories) for a bespoke response. Possible diagnoses are returned ranked by probability. The care plan content is generated using Claude and customized based on the patient's demographics.

Challenges we ran into

• Balancing speed vs accuracy in searching thousands of medical records
• Presenting diagnosis probabilities without overconfidence or liability
  
• Generating care plans that are tailored and thorough but concise

What we learned

• The tradeoffs between precision and speed for semantic search
• The difficulties of parsing medical terminology into lay explanations
• The potential for AI to enhance (not replace) clinician expertise

What's next for Aida

• Integrating into NHS digital health systems
• Incorporating patient feedback into care plan content
• Expanding databases for rare diseases and personalized medicine

We're excited to continue refining Aida to help NHS providers efficiently deliver quality care, improve patient understanding, and potentially save millions in taxpayer money.

Built With

• claude
• langchain
• weaviate