At Turing, I evaluated frontier LLM responses for correctness, reasoning quality, and instruction-following, producing preference signals for RLHF reward modeling. I also authored reference solutions and curated SFT datasets with adversarial and edge-case prompts.
For Friday, I built a persistent fleet of five specialist AI agents that generates evidence-backed commerce strategy recommendations. I developed its FastAPI modular monolith and tested it across parallel missions, budget limits, and cancellation and timeout paths.
I also built Anton, a PDF-grounded RAG agent with pgvector retrieval and a two-runtime backend, and ForkedUp, a video-first food discovery and ordering platform. I'm developing Doable, an AI-powered app builder that generates, scaffolds, and deploys applications from user descriptions.
