Software engineer
building cloud infrastructure at scale.

Amazon Web Services

• Lead design for complex features enabling customers to bridge security compliance gaps and expand customization — spanning task state management, cross-service communication, security changes, and a public API.
• Lead expansion of the service to new AWS regions across commercial and ADC partitions; serve as subject matter expert for region builds within the team.
• Ran comprehensive load tests to scale the service from a 15,000 to 50,000 account limit, eliminating stuck operations and recurring service alarms.
• Own the full project lifecycle — design, implementation, testing, documentation, deployment — across all AWS regions.
• Reduced on-call load by ~25–30% by designing a new locking mechanism in a serverless service to resolve a race condition.

NCR Corporation

• Built a microservice integrating internal services with Google Business APIs using Loopback, OpenAPI specs, GCP SQL/Scheduler, and an Angular frontend.
• Connected the microservice to an existing .NET codebase — frontend in JavaScript (ExtJS), backend in C#.
• Documented APIs with Swagger and managed deployment with Kubernetes and Argo CI/CD.

Openwater VC

• Developed a React Native application for the startup Saymile.
• Built a Node.js/Express backend with Firebase supporting messaging, group ordering, order tracking, and phone number authentication.

AI Website Auditor

A user submits a website URL. The system screenshots the rendered page with headless Chromium, sends the image to a vision LLM for a usability critique, and stores the result for retrieval — all on an asynchronous, serverless pipeline. Tap a component to see its role.

Stack: API Gateway · Lambda (Node.js) · DynamoDB Streams · SQS · S3 · headless Chromium · vision LLM. The design is fully event-driven: writing a row to DynamoDB emits a stream event that drives the rest of the pipeline — no polling. A router consumes the stream and hands off to SQS, so the slow screenshot-and-audit work (a 30+ second browser render) runs on the queue with its own retries and back-pressure, never blocking the stream shard iterator or the API response.

Try it live Demo offline

Enter a public website URL and the pipeline will screenshot it and return an AI usability critique. Takes ~30–60 seconds end to end.

Audit Clear

Queued Screenshot Analyzing Done

The live demo is switched off while backend rate limits and spend caps are being finalized. The architecture above is fully built and deployed.

Stadium IOPT

• Managed a team of frontend and backend developers building a Swift iOS application with minigames to increase mid-game fan interaction and safety at Bobby Dodd Stadium.

Independent Research

• Created a demo for the paper "Feel The Music: Automatically Generating A Dance For An Input Song" with Professor Devi Parikh — using MFCC audio features and Markov states with beam search to generate dance videos from songs.

Deployed on AWS

The whole stack is declared in a single CloudFormation template; content updates ship via a shell script that syncs S3 and invalidates the CloudFront cache. Tap a component below to see what it does.

Typeset in Fraunces & Inter. Built by hand, deployed by template.