Building a serverless AI assistant at Pelago: concept to care in two weeks

This article describes how Pelago, a digital health company, built a serverless AI assistant for substance use disorder coaches in just two weeks using AWS services.

• Event-driven architecture uses SNS to fan out messages to independent Lambda functions for decoupled processing
• Chat Assistant Lambda retrieves full conversation history from DynamoDB and invokes Amazon Bedrock asynchronously to generate contextual suggestions
• Pre-generated suggestions are stored in MySQL and retrieved in under 100ms when coaches open conversations
• VPC endpoints keep Protected Health Information (PHI) off the public internet, maintaining HIPAA compliance
• Serverless architecture automatically scales with traffic spikes and eliminates idle compute costs during off-peak hours
• System reduced response preparation time by 40% with 79.6% of AI suggestions rated helpful by care team

The solution demonstrates how managed AI services and serverless patterns enable rapid deployment of compliant healthcare applications without infrastructure overhead.