Guided Series

Eight field guides adapted from Distilled: The Engineering of Small, Fast, Cheap AI Models, from transfer signals and synthetic curricula to release pipelines and specialist fleets.

A source-guided course on Kafka protocol internals, platform ownership, KRaft and KIP-848 migrations, compatibility, portability, performance, and production failure boundaries.

Eight essays on data-movement ownership across Linux networking, storage, kernel bypass, RDMA, accelerators, serialization, and the failure modes that make copying cheaper.

A journey from over-engineered build-time rendering to elegant client-side solutions. Explores three architectures and learns valuable lessons about premature optimization.

Master efficient LLMs under 3B parameters—models that match 5× larger competitors on reasoning tasks. From distillation to edge deployment, learn the techniques making AI accessible everywhere.

Four hands-on workshops for taking any model release from artifact audit to durable agents, long-context storage, and topology-aware multi-GPU inference.

A source-guided engineering study of Kimi K3 plus a lab track that turns its architecture, API, optimizer, routing, and memory ideas into reproducible experiments.

Comprehensive technical analysis of Andrej Karpathy's nanochat project. Deep dives into architecture, training pipeline, optimizers, and proposals for evolving this minimal LLM implementation.

An eight-part source-guided study of Redpanda 26.1.13: Kafka compatibility, Raft durability, storage, remote data, transforms, RPC, cluster control, and benchmark design.