At Freedom Music, I designed and built a recommendation system for a music-streaming catalog of approximately 60 million tracks, ranking recommendations with audio embeddings. I also built a personalized infinite radio stream that adapts track selection in real time to user feedback.
I built the serving layer around nightly batch jobs for user taste profiles and track embeddings, with vector search and scoring on the request path. I also set up CI/CD quality gates and containerized deployments to Kubernetes across development and production.
At WONK, I built agent workflows for an edtech platform that generates lesson plans for teachers. I developed tools and API endpoints, and implemented session context, long-term memory, and personalization for generated content.
At TSARKA, I developed internal ML workflows using local LLM inference and context retrieval to generate client-facing case descriptions from operational reports. I also prototyped a sandboxed security-audit agent and clustered approximately 500 SIEM detection rules by embedding similarity.
