LLM EngineeringLab

Understand how modern LLM systems really work.

27 interactive modules covering inference optimization, attention architectures, retrieval, fine-tuning, privacy and evaluation. Each one uses analytical models, exact computations or deterministic simulations, and every result says which. Connect Ollama, an OpenAI-compatible API or Hugging Face to run real models.

KV & prefix cachingSpeculative & Medusa decoding RoPE · Sinks · Ring · MoELoRA · QLoRA · Quantization Hybrid search · Reranking · RAGDP-SGD · Guardrails · Calibration