Backend Software Engineer — Golang · AWS · Distributed Systems · AI/ML Tooling
Building production-hardened backend systems and the AI infrastructure that accelerates how engineers work.
Experience
Capital One · Software Engineer · 2024–Present
The Modern Account Writes APIs
Capital One's first Credit Card Accounts Writes API modernized from SOAP to REST API. I owned it end-to-end from day one: Golang application logic, OAuth2 security, AWS CDK infrastructure, OpenTelemetry observability, component testing, and K6 performance testing. Zero to production-hardened in 4 months.
The Fargate Migration
Lambda couldn't meet our SLAs. I benchmarked both environments and demonstrated 5–10× better P95 latency on Fargate, built the CDK infrastructure with distroless containers and graceful shutdown logic, and presented the findings to leadership. The result: an org-wide architecture decision and an unblocked service.
The Memory Leak
Containers were crashing under sustained load during endurance testing. I used pprof to profile 10,000+ concurrent requests, traced the leak to unclosed response bodies in nested logging middleware, and eliminated OOM crashes entirely.
Performance Testing Architecture
Our API had zero performance test coverage and no clear path to get there — K6 is Capital One's mandated load testing tool, but K6 and Golang are natively incompatible runtimes. I evaluated four architectural patterns, designed a bridging solution between the two, and ported AES-GCM encryption with verified cipher parity so auth behaved identically under load. Engineers can now run complex load tests in under an hour.
The AI Tooling Repository
Five modernization teams, no standardized way to build or share AI tools. I socialized the idea, built the repository structure, wrote the symlink integration system, and got org-wide adoption. A contributed multi-agent workflow now cuts API modernization effort from 2 weeks to half a day.
The Pipeline Debug
A silent Docker artifact-mapping failure in Capital One's central deployment pipeline was blocking teams across the entire tech tower. I diagnosed it from source code alone, engineered a workaround, and shared findings that unblocked multiple teams.
Research
Optimized a financial market order book simulation by migrating NumPy/Pandas data structures to PyTorch, improving computational throughput for high-frequency trading research.
Built an automated OCR pipeline digitizing 4,000+ noisy historical financial articles using Tesseract, docTr, and EasyOCR; benchmarked 15B–70B parameter LLMs (Llama, Command-R, Mixtral) to support historical financial dataset digitization.