About
I’m Necdet Duruk, a Senior ML Engineer based in Mississauga, Ontario.
For the past six years I’ve built production machine learning systems for time-series problems — anomaly detection on telecom network telemetry, forecasting, real-time scoring pipelines. The work runs at scale on data that doesn’t behave the way textbook datasets do, which has shaped most of how I think about ML.
I write here about the parts of ML engineering that don’t get enough attention: rigorous evaluation, the gap between published benchmarks and production reality, foundation models meeting traditional time-series workloads, and the engineering practices that separate projects that ship from ones that don’t.
What I’m working on
- Benchmarking time-series foundation models — Chronos, TimesFM, Moment — against well-tuned classical baselines on real production telemetry. The project is public and active on GitHub.
- Writing up findings as they come in. Part 1 is live; part 2 (foundation models) is in progress.