Ed N.

Staff Applied Scientist @The Trade Desk

London, GB

Leading our AI forecasting model infra development. Designed, built and optimized our model training and inference stack. The main focus is to make it quick and easy to get models into production that provide customer facing metrics in real-time.Building scalable infra that runs distributed GPU training on millions of datapoints daily and handles millions of requests a week performing multiple feature transformations and inferences for transformer architecture models in ~20ms.Driving significant accuracy improvements aiding better decisions, resulting in better outcomes for our clients.Reduced inference latency by 95+% through a combination of better de/serialization and python code optimization. This also decreased training data generation compute cost by 40%.Raising coding standards within the team and beyond by setting best practices, adding linting and type checking to CI/CD as well as providing high quality code reviewsTech Used/Buzzwords: Python, Databricks, AWS, PyTorch, Triton, TorchServe, KServe, Spark, PySpark, Ray, MLFlow, Airflow, FastAPI, asyncio, pydantic, Ruff, ty

University of Leicester

Master of Mathematics (MMath), Mathematics

Central Sussex College Haywards Heath

A Levels, Maths, Further Maths, Computing, Chemistry