Improving the inference-time performance of the Gemini family models.

• Optimised inference performance of Gemini 2-3 models across all model sizes. Contributed inference optimisations for the IMO gold model https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/

• Core contributor to long-context (1M tokens) infrastructure and distributed computing partitioning strategies.

• My contributions directly reduce latency and compute usage for models running in prod, on maaaany TPUs.

Areas of expertise – distributed systems, LLM scaling and partitioning strategies, TPU architectures, JAX.

Teaching weekly Discrete Maths, Logic, and Algorithms tutorials to a group of 10 first-year students.