HomeDashboardPricingSettings
Learning spaces
Python with AIPython fundamentals through FastAPI, Pydantic, agents, MCP and Langfuse: build a real AI backend end to end.5 free lessonsInference EngineeringHow LLMs and generative models are served in production: GPUs, KV cache, batching, quantization, speculation, parallelism and autoscaling.5 free lessonsAI ResearchA structured path from math fundamentals to LLMs, research papers, tooling, and applied practice.16 free lessonsSystem Design In DepthDeep-dive notes, case studies, and from-scratch implementations for senior engineering interviews.5 free reads
Sign in

All lessons

40 pages · 5 free · 35 Pro

Read the free lessons in full. Pro unlocks every remaining lesson and its video.

IE Orientation2Video lessons
4:14

What inference engineering is

Lesson 01 4:14Watch
4:26

The life of an inference request

Lesson 02 4:26Watch
01 Before You Optimize4Video lessons
4:38

Shared APIs versus dedicated deployments

Lesson 01 4:38Watch
4:24

Latency budgets and workload shapes

Lesson 02 4:24 PRO
4:54

Measuring TTFT, TPS, and percentiles

Lesson 03 4:54Watch
4:26

Picking the smallest model that passes

Lesson 04 4:26 PRO
02 How Models Run7Video lessons
5:21

Matmul and activations for inference engineers

Lesson 01 5:21 PRO
5:13

Tokens, chat templates, and sampling

Lesson 02 5:13 PRO
4:48

Prefill and decode

Lesson 03 4:48Watch
4:45

Transformer blocks and attention

Lesson 04 4:45 PRO
4:52

The KV cache and its memory bill

Lesson 05 4:52 PRO
4:52

Mixture of experts models

Lesson 06 4:52 PRO
4:48

How image and video models generate

Lesson 07 4:48 PRO
03 Bottlenecks and Hardware5Video lessons
4:30

Compute-bound or memory-bound

Lesson 01 4:30 PRO
4:41

Inside a datacenter GPU

Lesson 02 4:41 PRO
5:07

Choosing a GPU for a workload

Lesson 03 5:07 PRO
4:55

Nodes, interconnects, and MIG

Lesson 04 4:55 PRO
4:21

Making attention cheaper

Lesson 05 4:21 PRO
04 The Software Stack5Video lessons
4:49

CUDA kernels, selection, and fusion

Lesson 01 4:49 PRO
4:54

PyTorch, compile, and model files

Lesson 02 4:54 PRO
5:13

vLLM, SGLang, and TensorRT-LLM

Lesson 03 5:13 PRO
5:14

Continuous batching

Lesson 04 5:14 PRO
4:52

Benchmarking and profiling inference

Lesson 05 4:52 PRO
05 Quantization and Speculation4Video lessons
5:31

Number formats for inference

Lesson 01 5:31 PRO
5:17

Quantizing without losing quality

Lesson 02 5:17 PRO
4:23

Speculative decoding

Lesson 03 4:23 PRO
5:09

EAGLE, Medusa, and n-gram drafts

Lesson 04 5:09 PRO
06 Caching, Parallelism, and Disaggregation4Video lessons
4:35

Prefix caching

Lesson 01 4:35 PRO
5:00

KV cache tiers and cache-aware routing

Lesson 02 5:00 PRO
5:20

Tensor, expert, and pipeline parallelism

Lesson 03 5:20 PRO
5:14

Disaggregated prefill and decode

Lesson 04 5:14 PRO
07 Beyond Text3Video lessons
5:23

Vision language and embedding models

Lesson 01 5:23 PRO
4:53

Speech pipelines: ASR and TTS

Lesson 02 4:53 PRO
5:19

Faster image and video generation

Lesson 03 5:19 PRO
08 Inference in Production6Video lessons
5:09

Containers and cold starts

Lesson 01 5:09 PRO
5:06

Autoscaling, routing, and queues

Lesson 02 5:06 PRO
4:41

Multi-cloud capacity and reliability

Lesson 03 4:41 PRO
4:59

Shipping changes, cost, and observability

Lesson 04 4:59 PRO
5:01

Client code, streaming, and protocols

Lesson 05 5:01 PRO
5:14

Capstone: an inference stack for Kite

Lesson 06 5:14 PRO