This prototype is a simulation platform for pharmacy education built on AWS. Students interact with AI-powered patient personas through text and voice chat, practicing clinical assessment and communication skills in a safe, repeatable environment. Instructors create simulation groups with configurable patient scenarios, and the system generates AI-powered debriefs evaluating student performance against key clinical questions.
| Document | Description |
|---|---|
| Deployment Guide | End-to-end deployment instructions, prerequisites, and troubleshooting |
| Architecture Deep Dive | System architecture, component breakdown, data flows, and database schema |
| User Guide | How to use the platform as a student, instructor, or admin |
| Modification Guide | Customizing colors, extending the API, configuring LLMs, and local dev setup |
| Database Migrations | Migration system usage, patterns, and safety guidelines |
| Dependency Management | Dependency locking and update procedures for Python and Node.js |
| API Documentation | REST API endpoints, request/response formats |
| Data Ingestion | Document processing pipeline for patient case materials |
| Bedrock Guardrails | AI safety guardrails configuration |
| Voice Agent Setup | Bedrock AgentCore voice agent deployment |
| Voice Agent Deep Dive | Voice architecture and Nova Sonic integration |
| Custom Domain & SES | Custom domain setup, SES email delivery, and Amplify hosting |
| Contributing to Docs | Guidelines for writing and maintaining documentation |
| Changelog | Version history and release notes |
For a detailed breakdown of each component, data flows, and the full database schema, see the Architecture Deep Dive.
.
├── cdk/ # AWS CDK infrastructure + backend services
│ ├── bin/ # CDK app entry point
│ ├── lib/ # CDK stack definitions
│ ├── lambda/ # Lambda function source code (JS + Python)
│ ├── text_generation/ # AI text generation service (Docker Lambda)
│ ├── data_ingestion/ # Document processing pipeline (Docker Lambda)
│ ├── socket-server/ # Real-time server (ECS Fargate)
│ ├── voice-agent/ # Bedrock AgentCore voice agent
│ └── layers/ # Lambda layers (shared dependencies)
├── frontend/ # React SPA (Vite + Tailwind + shadcn/ui)
│ └── src/
│ ├── pages/ # Route-level page components
│ ├── components/ # Reusable UI components
│ ├── services/ # API service layers
│ ├── hooks/ # Custom React hooks
│ └── lib/ # Shared utilities
├── docs/ # Project documentation
└── scripts/ # Utility scripts
| Directory | Description |
|---|---|
/cdk |
AWS CDK infrastructure code (TypeScript) |
/cdk/bin |
CDK app entry point and stack orchestration |
/cdk/lib |
CDK stack definitions for all infrastructure (VPC, database, API, auth, hosting, WAF) |
/cdk/lambda |
Lambda function source code (Node.js 22.x and Python 3.12) |
/cdk/layers |
Lambda layers for shared dependencies (postgres, psycopg2, aws-jwt-verify, jose) |
/cdk/text_generation |
AI text generation service — LangChain + Bedrock for chat, debriefs, question matching (Docker Lambda) |
/cdk/data_ingestion |
Document processing pipeline — PDF ingestion into pgvector store (Docker Lambda) |
/cdk/socket-server |
Real-time server — Socket.IO voice on ECS Fargate |
/cdk/voice-agent |
Bedrock AgentCore voice agent with Nova Sonic |
/frontend |
React 19 frontend application (TypeScript, Vite, Tailwind CSS, shadcn/ui) |
/frontend/src/pages |
Page-level components organized by user role (student, instructor, admin) |
/frontend/src/components |
Reusable UI components and shadcn/ui primitives |
/frontend/src/services |
API service layer for backend communication |
/frontend/src/hooks |
Custom React hooks including WebSocket and panel management |
/frontend/src/lib |
Shared utilities (auth, API client, browser detection, debrief parsing) |
/docs |
Project documentation including deployment, architecture, and user guides |
- React 19 with TypeScript
- Vite 7 for build tooling
- Tailwind CSS 4 for styling
- shadcn/ui (Radix UI) for UI components
- AWS Amplify v6 for Cognito authentication
- Recharts for analytics charts
- React Router v7 for client-side routing
- Socket.IO client for real-time communication
- AWS Lambda (Node.js 22 and Python 3.12) for serverless compute
- Amazon Bedrock with Claude Sonnet 4.6 (Anthropic) for LLM inference
- Cohere Embed v4 via Amazon Bedrock for vector embeddings
- LangChain for AI orchestration (chat, debriefs, question matching)
- PostgreSQL 16 (Amazon RDS) with pgvector for relational and vector data storage
- Amazon DynamoDB for conversation session state
- Amazon S3 for document and file storage
- Amazon API Gateway (REST) for API management
- AWS AppSync (GraphQL) for real-time text streaming subscriptions
- Amazon Cognito for authentication and authorization
- Socket.IO on ECS Fargate for WebSocket communication
- Amazon Bedrock AgentCore with Nova Sonic 2.0 for speech-to-speech voice AI
- AWS CDK v2 (TypeScript) for infrastructure as code
- AWS CodePipeline and CodeBuild for CI/CD
- Amazon ECS Fargate for containerized services
- Amazon RDS with RDS Proxy for managed PostgreSQL
- Amazon VPC for network isolation
- AWS WAF for API and AppSync protection
- Amazon CloudFront for content delivery
- Amazon SES for email delivery
- AWS Amplify Hosting for frontend deployment
This application was architected and developed by Ayush Srihari and Rajrupa Sanyal with project assistance by Carrie Schulz. Thanks to the UBC Cloud Innovation Centre Technical and Project Management teams for their guidance and support.
This project is licensed under the MIT License.
Licenses of libraries and tools used by the system are listed below:
PostgreSQL license - For PostgreSQL — a liberal open source license, similar to BSD or MIT.
Anthropic Acceptable Use Policy - For Cohere Embed English v3, accessed via Amazon Bedrock for vector embeddings.
Cohere terms of use - For Claude Haiku 4.5 and Claude Sonnet 4.6, accessed via Amazon Bedrock for text generation.