Free & Open Source20 LessonsReal Code

AI Engineering from First Principles

Go from zero to production AI systems. Every technique you learn, you prove with benchmarks. No fluff, no theory-only lectures. Working code with SOTA results.

5
Levels
20
Lessons
50+
Code Examples
SOTA
Benchmarks
Start LearningView Roadmap

What You'll Build

Level 2

Voice Assistant

Whisper + LLM + TTS. 2-second response time.

Level 3

Production RAG

Hybrid search + reranking. RAGAS score >0.8.

Level 4

AI Agent Pipeline

Tool calling + multi-agent orchestration.

Learning Roadmap

0

Foundations

No AI experience required

0.1

What is an Embedding?

Vectors that capture meaning. Interactive visualization.

0.2

APIs vs Local Models

Cloud or self-hosted? Cost calculator included.

0.3

Tokens & Context Windows

How LLMs read text. Pricing per million tokens.

0.4

Environment Setup

Python, API keys, GPU setup, Docker.

1

Single Blocks

Master individual building blocks

1.1

Text Embeddings

MTEB benchmarks. BGE, OpenAI, Cohere comparison.

1.2

Image Search with CLIP

Zero-shot classification. Cross-modal search.

1.3

Speech Recognition

Whisper. WER benchmarks. Speaker diarization.

1.4

Text Classification

Zero-shot vs fine-tuned. GLUE/SST-2 benchmarks.

1.5

Object Detection

YOLO v11, RT-DETR, Grounding DINO. COCO mAP.

1.6

Image Segmentation

SAM 2. Point prompts, automatic masks.

1.7

Text-to-Speech

OpenAI TTS, ElevenLabs, Coqui, Bark.

1.8

Document Parsing

Docling, Marker, VLM OCR. OmniDocBench.

2

Pipelines

Chain 2-3 blocks together

2.1

Caption + Search

Image captioning + embeddings for visual search.

2.2

Voice Assistant

Whisper → LLM → TTS. Latency optimization.

2.4

Basic RAG

Chunk, embed, retrieve, generate.

3

Production

Real-world patterns

3.1

Hybrid Search

BM25 + vector. Reciprocal Rank Fusion.

3.2

Reranking

Cross-encoder reranking. NDCG improvements.

3.3

Document RAG

Chunking strategies, citations, RAGAS eval.

3.4

Hallucination Detection

NLI verification, self-consistency, faithfulness.

4

Advanced

Complex compositions

4.1

Multi-modal RAG

RAG over images, tables, and text. ColPali.

4.2

Agent Pipelines

ReAct, tool calling, multi-agent orchestration.

4.3

Video Understanding

Frame sampling, Video-LLMs, temporal analysis.

4.4

Real-time Systems

Streaming, vLLM, latency optimization.

Why This Course

Other Courses

  • "Week 1: Introduction to AI" (calendar-locked)
  • Toy examples that don't work in production
  • "Trust us, this works" - no benchmarks
  • Outdated by the time you finish

This Course

  • Self-paced. All 20 lessons available now.
  • Production code you can copy-paste
  • Every technique linked to SOTA benchmarks
  • Updated with latest models and techniques

Start Now

Begin with embeddings. Build up to production RAG and agent pipelines.

Start Lesson 0.1Explore Building Blocks

Deep Dives

Standalone technical deep-dives into foundational ML concepts. No prerequisites — read in any order.

DEEP DIVE

How Transformers Work

Attention mechanism, positional encoding, and the architecture behind every modern LLM.

DEEP DIVE

Embedding Dimensions

Why 768? Why 1536? The math and tradeoffs behind embedding vector sizes.

DEEP DIVE

Matrix Operations

The linear algebra that powers neural networks. Dot products, projections, and attention as matrix math.