AI Medical Assistant

AI Medical Assistant — Streamlining Clinical Workflows with Generative AI 🚀 Inspiration

Healthcare professionals spend a significant portion of their time collecting patient details, summarizing symptoms, and preparing medical reports. This repetitive process delays diagnosis and reduces face-to-face time with patients. We were inspired to create an AI Medical Assistant that automates early-stage symptom analysis, retrieves relevant medical knowledge, and generates structured reports — helping doctors focus on clinical decisions rather than documentation.

🧠 What We Built!

We built a Streamlit-based conversational medical assistant powered by AWS Bedrock, OpenSearch, and LangChain, which can:

Conduct interactive conversations with users to understand symptoms.

Summarize medical history using a Chat Summary Agent.

Retrieve relevant clinical information using OpenSearch for RAG-based retrieval.

Generate structured, doctor-friendly reports through a Report Generator Agent.

Incorporate doctor feedback loops to continually improve AI responses.

🏗️ Architecture Overview

Our architecture includes multiple specialized agents:

Conversation Agent – Handles patient dialogue via natural language.

Chat Summary Agent – Extracts and summarizes user responses for clarity.

Retrieval Agent – Connects to an OpenSearch Knowledge Base for context-aware retrieval.

Report Generator Agent – Produces a structured clinical summary and recommendation draft.

Feedback Loop – Updates and fine-tunes responses based on doctor edits.

All these components are orchestrated through Streamlit UI, backed by AWS Bedrock models for LLM and embeddings.

⚙️ How We Built It

Frontend: Streamlit for an intuitive conversational interface.

LLM Backbone: AWS Bedrock (Claude by Anthropic) for reasoning and summarization.

Retrieval: OpenSearch used directly for efficient semantic search.

Deployment: Dockerized application pushed to Amazon ECR, deployed on ECS (Fargate) for scalability.

💡 What We Learned

How to integrate AWS Bedrock seamlessly with LangChain for both embeddings and LLM inference.

Efficient retrieval-augmented generation (RAG) implementation using OpenSearch.

Best practices for containerization, versioning, and deployment in ECS/ECR.

Building modular AI agents that work collaboratively in a single workflow.

🧩 Challenges Faced

Handling validation errors during Bedrock model invocation (e.g., ensuring valid message structures).

Managing environment files (.env) and dependencies during Docker builds.

Designing a multi-agent pipeline that remains stateless yet contextually coherent.

Fine-tuning prompt engineering for accurate symptom-to-report conversion.

🌟 Future Enhancements

Integrate voice-based symptom collection using AWS Transcribe.

Add multilingual support for diverse patient populations.

Implement doctor dashboard analytics for insights into case patterns.

Fine-tune a domain-specific medical LLM using de-identified datasets.

❤️ Impact

By combining the power of AWS AI services and open-source retrieval, the AI Medical Assistant demonstrates how AI can reduce clinical workload, improve patient interaction quality, and enable faster, data-driven medical insights.