Overview
A leading healthcare provider needed a comprehensive, AI-powered platform to streamline patient interactions, automate document processing, and deliver personalized treatment recommendations. We designed and built a full-stack solution on AWS serverless infrastructure — combining conversational AI agents, intelligent document processing, and an evidence-based recommendation engine into a single integrated system.
The platform serves both providers and patients, with a focus on diabetes care and management. Every component was built with HIPAA-aware architecture: encryption at rest and in transit, audit logging, PHI sanitization, and role-based access controls.
Three Specialized AI Engines
1. Medical Chat Engine
Conversational AI for patient interactions, medical queries, and health management. Built on the Strands Agents framework with 11 specialized tools including diabetes and AMD specialist consultations, medication management, appointment scheduling, and exercise tracking.
- Privacy-safe personal health tools (no patient ID exposure)
- Knowledge base integration with medical guidelines
- Emergency keyword detection and routing
- Streaming responses with conversation memory persistence
2. Intelligent Document Processing Engine
Automated extraction of medical information from intake forms, prescriptions, insurance cards, and doctor’s notes. Uses a multi-model strategy — lightweight orchestrator for routing plus vision-capable model for complex document layouts and handwriting.
- Automatic document classification and structured data extraction
- Handles structured, semi-structured, and unstructured documents
- PHI sanitization and guardrail-based content filtering
- Processing: upload → classify → extract → validate → persist
3. Personalized Recommendation Engine
Generates evidence-based health recommendations using patient data, medical knowledge bases, and AI. Covers exercise, diet, medication, lifestyle, and monitoring — each with actionable steps, rationale, expected outcomes, and safety considerations.
- Five recommendation types with confidence scoring
- Patient context from demographics, vitals, HbA1c, medications
- Knowledge base integration with clinical research
- Status tracking: pending → accepted → completed
Architecture & Infrastructure
The platform uses a fully serverless architecture across six layers — from the React frontend distributed via CDN, through JWT-based authentication, REST APIs, AI agent runtimes with microVM isolation, to an auto-scaling PostgreSQL database.
Frontend
React 18 + TypeScript, CloudFront CDN, Cognito authentication, Cloudscape Design System
AI Layer
Amazon Bedrock AgentCore with microVM isolation, Strands Agents framework, multi-model strategy
Backend
API Gateway, Lambda functions, serverless compute, one-command CDK deployment
Data & Security
Aurora PostgreSQL Serverless v2, KMS encryption, audit logging, HIPAA-aware access controls
Performance
Simple Queries
2–5 seconds
Complex Consultations
5–15 seconds
Document Processing
10–30 seconds
Database Queries
<100ms
Interested in a similar solution for your organization?
Request a Consultation