Free, visual, interactive guide to AI - covers Deep Learning, LLMs, the Transformer, Vector Databases, RAG, and Agents. By Rajul Babel.
28 sections grouped into 6 super-sections. Each chapter uses animations, interactive diagrams, and step-by-step breakdowns to build intuition before showing the math.
- Foundations of Deep Learning - neuron, layer, weights/biases, activations, forward/backward pass, gradient descent, dropout, Adam, weight init, linear algebra, training deep networks.
- The Rise of LLMs - tokenization, self-supervised learning, cross-entropy, SFT, RLHF, scaling laws, distillation, contrastive learning.
- The Transformer Era - CNN/RNN motivation, embeddings, positional encoding (sinusoidal & RoPE), Q/K/V intuition, scaled dot-product attention, multi-head, encoder (Add & Norm, FFN), decoder (causal masking, cross-attention), KV cache, grouped-query attention, mixture of experts, reasoning models.
- Vector Databases at Depth - distance metrics, IVF, HNSW, Vamana, scalar/product/binary quantization, Matryoshka, IVF-PQ, HNSW+PQ, filtering, sharding, hybrid search, rerankers, FAISS, pgvector, Qdrant, Pinecone, Weaviate, Milvus, Chroma.
- Retrieval Augmented Generation (RAG) - naive pipeline and failure modes, ingestion + chunking, embedding choice and domain adaptation, hybrid retrieval, reranker cascade, HyDE, multi-query, query routing, context packing, citations, multi-hop, Self-RAG, CRAG, GraphRAG, agentic RAG, eval (RAGAS, golden datasets), caching, cost, observability.
- Agentic AI - prompt anatomy, system prompts, few-shot, CoT, tool use, JSON schemas, MCP architecture and primitives, A2A, agent loops (ReAct, plan-execute), memory types, multi-agent (orchestrator-worker, supervisor, critic/debate), evals, guardrails, injection defenses, framework picks.
- React 18 - hooks-only component tree
- Vite - build toolchain
- GitHub Actions - CI/CD
- GitHub Pages - hosting
npm install
npm run devOpen http://localhost:5173/learn-ai/
npm run build
npm run previewPushes to main automatically build and deploy via GitHub Actions.
Rajul Babel - LinkedIn - GitHub
MIT