Lectures
Machine Learning: From Data to Large Language Models
An ~80–90 min introduction (no prior ML needed): fundamentals, problem types, production ML systems, embeddings & Transformers, using LLMs (RAG, agents, local models), shipping ML in a real platform, ethics & privacy — ending in a hands-on capstone case study.
Open slides →The Infrastructure Behind AI
The systems view of AI: what it takes to actually run a model. Why the GPU isn't a CPU, why VRAM (and the KV cache) is the wall, fitting models with quantization / GQA / MoE, going fast with Tensor Cores & FlashAttention, serving many users with vLLM, scaling across GPUs and the network, cost per token, and sharing one cluster — tied back to the datacenter, cloud, and network lectures. Every number measured on a real RTX 3090.
Open slides →Generative AI: Reasoning Models & Agents reserve
A follow-on lecture: models that think before answering — chain-of-thought, test-time compute, and how reasoning is trained with reinforcement learning from verifiable rewards (RLVR / GRPO, DeepSeek-R1) — then how reasoning + tools + a loop becomes an agent, and how fast agent capability is growing.
Open slides →