creators

Umar Jamil - LLMs from scratch, Mistral AI engineer

Ingegnere di machine learning presso Mistral AI, insegna i transformer e gli LLM a partire dai principi fondamentali

In inglese

Umar Jamil is a machine learning engineer from Milan who works at Mistral AI. In March 2026 he announced the launch of Mistral's Forge platform, which lets organisations train specialised models on their own data, and was hiring reinforcement-learning infrastructure and research engineers to build it. He speaks Italian, English, Urdu and Mandarin. On YouTube he posts code-along lectures that derive a method and then implement it from scratch in PyTorch or Triton: a Transformer, Flash Attention, a vision-language model, DPO, RLHF, Mamba and S4, and a distributed training framework built from first principles. Twenty-five videos have drawn about 2.6 million views and 90,800 subscribers.

Specialization

transformers from scratch, LLM internals, GPU kernels and Triton, applied machine learning