Machine Learning Engineer
Hi, I'm Jacob.
I'm a machine learning engineer working on machine-learning systems and RL — training and serving large models, GPU kernels, and the infrastructure around them. I like understanding systems from first principles and writing the explanation I wish I'd had.
Explore the library
A first-principles reference library — seven areas, ~497 lesson pages across 23 tracks. Pick an area, or browse the full catalog.
Foundations
SICP in JavaScript, functional programming, classical ML, the deep-learning core, and computer vision — programming abstraction and typed effects through bias–variance, backprop, attention, detection, segmentation, and VLMs.
Open area →Generative models
Diffusion, flow matching, DiT, and tokenizers, plus a GPT built end-to-end from pretrain → SFT → CoT → DPO → RLVR.
Open area →Reinforcement learning
One linear track: MDPs → value & policy methods → TRPO/PPO → RLHF/GRPO → post-training systems → twenty applied domains.
Open area →GPU, kernels & serving
CUDA and Triton from first principles — including kernel-interview coding — the vLLM and SGLang serving engines, distributed training, and GenAI operations on Kubernetes.
Open area →Systems, data & design
Designing ML, Ray, distributed, agentic, and data-intensive systems end-to-end, and the data plane behind them.
Open area →Search, ads & recsys
Production ranking from first principles in three linear tracks: search (query understanding, BM25, dense & hybrid retrieval, learning-to-rank, reranking, relevance eval), recommender systems, and ads & auctions.
Open area →Model compression
Knowledge distillation from soft targets to on-policy and reasoning-model distillation, set against quantization and pruning.
Open area →✍️ Or read the Writing archive — earlier posts on algorithms, distributed systems, compilers, and databases.