Case Study + Deep Dive Telemedicine Support Agents with LangGraph/MCP - Dan Mason
AI Summary
The video is a detailed presentation by a software engineer from Stride, showcasing their work on building AI agent workflows for healthcare, specifically focusing on a telemedicine text-message-based system to assist patients during early pregnancy loss (miscarriage) treatment.
Key points from the video:
- Introduction to Stride’s custom software consultancy specializing in AI workflows and code generation.
- Case study highlighting the problem: manual button-pushing by humans for patient interaction, limiting scalability.
- The AI-powered system uses Langraph and Langchain frameworks with an LLM (Claude) at the core to flexibly manage patient communication and treatment workflows without coding for each new treatment.
- The system maintains a state for each patient, understanding treatment phase, timing, and local time zones, adapting dynamically to patient inputs.
- Hybrid human-in-the-loop design where the AI handles routine messaging with human supervisors reviewing complex or low-confidence cases.
- Use of blueprints (structured medical knowledge documents) to ensure medically approved language and safe interactions.
- Evaluation and confidence scoring mechanism to decide if AI responses need human approval.
- The system can handle multiple treatments and is deployed using a modern stack (Python, Node, React, MongoDB, Twilio, AWS).
- Challenges addressed include prompt injection safety, caching, and balancing cost versus performance.
- The approach enables roughly 10x scaling of patient support capacity compared to full manual operation.
- The team actively fine-tunes prompts and tooling to improve accuracy and maintainability without fine-tuning model weights.
- The system design emphasizes explainability, flexibility, and smooth human-AI collaboration.
Overall, the video provides an insightful example of how AI agent workflows can be effectively applied to healthcare telemedicine scenarios to increase scale, maintain safety, and improve operational efficiency.