I give technical talks that respect the audience's intelligence and their time — turning MLOps, LLM observability, and the messy reality of production ML into something people can use.
Making generative systems transparent and accountable — tracing, metrics, and evals that tell you what's actually happening.
Taking models from notebook to reliable production: lifecycle, tooling, and the unglamorous engineering that makes ML dependable.
Curation, annotation, and pipelines that don't rot — the data foundations everything else stands on.
Practical, honest guidance for engineers growing into ML — from first steps to standing on a conference stage.
A selection of recent sessions. Recordings and slides link out where available.
ML Conference · Munich, Germany November 27, 2025
Large language models fail in quiet, expensive ways. This talk makes the case that observability and rigorous evaluation aren't add-ons but the foundation of trustworthy LLM systems — how to instrument, trace, and hold generative models accountable so teams can actually improve what they ship.
ML Conference · Munich, Germany November 28, 2025
A hands-on workshop on building efficient, high-quality data pipelines for ML — using Voxel51 for dataset curation and visualization alongside Label Studio for annotation, and wiring them into a workflow teams can actually maintain.
ML Conference / devmio November 27, 2023
From experiment tracking to model registry and deployment: a practical look at using MLflow to bring order to the ML lifecycle, so models move from notebook to production without losing reproducibility along the way.
Evaluating RAGs & agents before they reach customers 2026
Speaker
Speaker
Photos from talks and workshops. (More going up as I gather them.)
One of those rare talks that made a genuinely hard topic feel obvious. I left with things I could apply the next morning.
Clear, energetic, and deeply technical without ever losing the room. Exactly the kind of speaker you want on the main stage.
Saumya turns MLOps from a buzzword into something you can actually build. Brilliant on stage and even better in the Q&A.