At Symmetry, I built a multi-provider LLM router for an AI-powered workout app, handling 1.65M generations per quarter for 231K users. It routes requests across Azure AI Foundry, AWS Bedrock, Gemini, and DeepSeek, with priority failover and per-request cost attribution.
I also migrated tables with up to 250M rows under live traffic using resumable backfills and non-blocking indexing. By adding functional indexes and removing N+1 queries, I cut search latency from 437 ms to 26.6 ms.
At Symmetry, I designed and shipped a ranking engine for 21M rank states, using EWMA with time decay and a weekly Celery pipeline. I migrated it from v1 to v2 through shadow comparison behind a kill switch.
On my part-time project, Skillion Studios, I design and operate the backend infrastructure as the sole backend engineer. The product has 16,000 monthly active users and approximately $14K ARR; I also developed offline-first sync and multiplayer features and implemented reliability and GDPR measures.
