Work

A few backend systems I've built and operated in production.

Event-Driven Notifications Platform

Migrated notifications and activity processing from a monolith to an event-driven serverless architecture.

Problem: Notifications and activity processing were tightly coupled inside a monolith, making them hard to scale and operate independently.
Approach: Re-architected the pipeline around events — decoupling producers and consumers with SNS and Kinesis, persisting state in DynamoDB, and processing asynchronously with Lambda.
Technologies: Node.jsAWS LambdaSNSKinesisDynamoDB
Impact: Scaled to over 1M events per day while maintaining sub-400ms response times, with clearer retries and ownership.
Lessons: Event-driven systems trade simplicity for scale — idempotency, retries, and observability have to be designed in, not bolted on.

Built backend infrastructure for configurable server-side A/B/n experiments.

Problem: Product teams needed to test backend-driven variants and measure impact without shipping bespoke logic each time.
Approach: Built a configurable experimentation framework for variant assignment and measurement handoff, usable across backend-driven product surfaces.
Technologies: Node.jsAWS
Impact: Enabled teams to test variants and analyze usage patterns to inform product decisions.
Lessons: Backend experimentation lives or dies on consistency — stable variant assignment and a clean measurement handoff matter more than clever logic.

Built and maintained payment-related product infrastructure for the giving platform.

Problem: The giving experience needed reliable payment flows across web and backend.
Approach: Built payment and giving flows integrating Stripe with web and backend systems, focused on reliability and a smooth user experience.
Technologies: StripeReactGatsbyNode.js
Impact: Delivered dependable payment flows for the giving platform.
Lessons: Payments demand correctness and clear failure handling above all — reliability is the feature.

Developed backend recommendation retrieval services using AWS Lambda and Redis caching.

Problem: Recommendations needed fast retrieval to keep product experiences responsive.
Approach: Built Lambda-based retrieval services backed by Redis caching to serve recommendations with low latency.
Technologies: AWS LambdaRedisNode.js
Impact: Improved response performance for personalized recommendations.
Lessons: Caching strategy is the heart of recommendation latency — the data path matters more than the model call.