I build end-to-end products — database schema to React component to the LLM pipeline behind it. My work sits at the seam between conventional software engineering and applied AI: where a model has to survive real traffic, real latency budgets, and users who don't care what a token is.
Currently: RAG and agent infrastructure for production workloads · Interested in: distributed inference, eval-driven development
| AI / ML |       |
| Frontend |     |
| Backend |      |
| Infra |     |
- Ship the boring version first. A working v1 with a dumb heuristic beats a clever pipeline that never lands.
- Evals before prompts. If you can't measure the output, you're not engineering it.
- Optimize the p95, not the demo. Users experience the slow path far more than the happy path.