Learn AI · Outline
Curriculum outline
One tree for everything: browse concepts in tracks, build projects on the guided path, and read real-world examples when you need production patterns. 6 tracks · 50 articles · 16 modules · 95 lessons · 11 examples.
Mental model: browse = conceptual atlas · guided = project spine · examples = industry-shaped writeups. Cross-links go both ways.
Learning progression
Browse tracks (full article list)
Concepts — Foundations → deeper theory
Core Concepts · 16 articles
- What is AI (and what it is not)
- Classical ML literacy · also in guided: Data splits and leakage
- Tokenization · also in guided: Tokenization and BPE
- Self-attention · also in guided: Self-attention
- Embeddings and similarity · also in guided: Embeddings and similarity
- Prompt engineering fundamentals · also in guided: Chat APIs and message roles
- Structured outputs · also in guided: Structured output
- RAG building blocks · also in guided: Embeddings and similarity
- Agents and the ReAct loop · also in guided: LLM vs agent; loops; stop conditions
- Evals fundamentals · also in guided: Offline vs online evals
- Serving and streaming · also in guided: API + streaming
- Networking for AI apps · also in guided: Retries, rate limits, idempotency
- LLM project lifecycle
- Multimodal basics · also in guided: Multimodal in → text
- Privacy and data for AI apps
- Open-weight models vs closed APIs · also in guided: Open-weight vs closed API
Advanced Concepts · 8 articles
- Scaling laws and compute · also in guided: Why your tiny LLM ≠ ChatGPT
- Fine-tuning with LoRA and QLoRA · also in guided: LoRA / QLoRA fine-tune
- Alignment basics — RLHF and DPO
- Mixture of Experts (MoE) · also in guided: Meta — LLM-scale training literacy
- Context engineering · also in guided: Context budgets
- Reasoning and test-time compute · also in guided: Reasoning models / test-time compute
- Multi-agent orchestration · also in guided: Workflow patterns
- Guardrails and safety systems · also in guided: Guardrails
Tools — Everyday stacks → production tooling
Key Tech · 8 articles
- LangGraph and LangChain patterns · also in guided: LLM vs agent; loops; stop conditions
- Hugging Face · also in guided: When SLM/FT beats RAG or prompting
- Ollama · also in guided: When SLM/FT beats RAG or prompting
- Model Context Protocol (MCP) · also in guided: MCP servers and clients
- LlamaIndex · also in guided: lab
- OpenAI and Anthropic APIs · also in guided: Chat APIs and message roles
- Chroma · also in guided: Vector DB
- Weights & Biases · also in guided: Eval vs base / teacher
Advanced Key Tech · 6 articles
- Ray · also in guided: API + streaming
- Triton Inference Server · also in guided: Netflix — in-house LLM serving
- Kafka for evented AI · also in guided: Workflow patterns
- OpenTelemetry for LLMs · also in guided: Tracing tool/LLM spans
- TensorRT-LLM and SGLang · also in guided: NVIDIA — agent toolkits and eval/obs
- MLflow for LLMOps · also in guided: Runbook
Data & inference — Retrieval data plane + serving
Data & Databases for AI · 6 articles
- Vector databases — what, why, and how · also in guided: Vector DB
- Choosing vector stores — Pinecone, Weaviate, Chroma, pgvector · also in guided: Vector DB
- Postgres and pgvector · also in guided: Vector DB
- Redis for AI caching · also in guided: Caching and latency
- Chunking and metadata · also in guided: Chunking
- Hybrid search and rerankers · also in guided: Hybrid search
Inference · 6 articles
- vLLM · also in guided: When SLM/FT beats RAG or prompting
- KV-cache, prefill, and decode · also in guided: Caching and latency
- Continuous batching · also in guided: API + streaming
- Quantization for inference · also in guided: Quantization tradeoffs
- Speculative decoding · also in guided: Serve your SLM
- Cost and latency routing · also in guided: Cost accounting
Guided path (every lesson)
Nothing here is “hidden” — every lesson is listed with browse concepts (and examples when mapped).
Onboarding — I — Foundations
Browse: LLM project lifecycle
- How this course works · browse: LLM project lifecycle
- Python, tooling, and compute options — Guided lab setup
- API keys, budgets, and responsible use · browse: Privacy and data for AI apps
- Rubric: tests, eval JSON, and README · browse: Evals fundamentals
What happened in AI (till now) — I — Foundations
Browse: What is AI (and what it is not) · Classical ML literacy · Open-weight models vs closed APIs
- Symbolic AI → why rules hit a wall · browse: What is AI (and what it is not)
- Classical ML era · browse: Classical ML literacy
- Deep learning boom · browse: Classical ML literacy
- Attention → Transformer → GPT lineage · browse: Self-attention, Tokenization
- Generative AI landscape · browse: What is AI (and what it is not), Multimodal basics
- Open-weight vs closed API · browse: Open-weight models vs closed APIs
- Jobs: AI eng, ML eng, research, FDE, PM · browse: LLM project lifecycle
ML/DL literacy — I — Foundations
Build an LLM from scratch — I — Foundations
Browse: Tokenization · Self-attention · Alignment basics — RLHF and DPO
- Tokenization and BPE · browse: Tokenization
- Next-token prediction · browse: Tokenization
- MLP language model · browse: Classical ML literacy
- Self-attention · browse: Self-attention
- Tiny GPT / mini-LLM end-to-end · browse: Self-attention, Scaling laws and compute
- Sampling (temperature, top-k/p) · browse: Prompt engineering fundamentals
- Pretrain vs SFT vs preference (map only) · browse: Alignment basics — RLHF and DPO, LLM project lifecycle
- Why your tiny LLM ≠ ChatGPT · browse: Scaling laws and compute
Talk to models in the real world — II — Generative AI practice
Browse: Prompt engineering fundamentals · Structured outputs · Multimodal basics · OpenAI and Anthropic APIs
- Chat APIs and message roles · browse: OpenAI and Anthropic APIs, Prompt engineering fundamentals
- Structured output · browse: Structured outputs
- Prompt versioning · browse: Prompt engineering fundamentals, Weights & Biases
- Multimodal in → text · browse: Multimodal basics
- Code assist loops · browse: Agents and the ReAct loop
RAG — II — Generative AI practice
Browse: RAG building blocks · Embeddings and similarity · Vector databases — what, why, and how · Chunking and metadata · Hybrid search and rerankers
- Embeddings and similarity · browse: Embeddings and similarity
- Chunking · browse: Chunking and metadata
- Vector DB · browse: Vector databases — what, why, and how, Choosing vector stores — Pinecone, Weaviate, Chroma, pgvector, Chroma
- Hybrid search · browse: Hybrid search and rerankers
- Rerank / rewrite · browse: Hybrid search and rerankers
- Citations and failures · browse: RAG building blocks, Guardrails and safety systems
Build & serve your SLM — II — Generative AI practice
Browse: Fine-tuning with LoRA and QLoRA · vLLM · Quantization for inference · Ollama · Hugging Face
- When SLM/FT beats RAG or prompting · browse: Fine-tuning with LoRA and QLoRA
- Dataset curation and licenses · browse: Privacy and data for AI apps
- LoRA / QLoRA fine-tune · browse: Fine-tuning with LoRA and QLoRA, Hugging Face
- Eval vs base / teacher · browse: Evals fundamentals
- Serve your SLM · browse: vLLM, Ollama, Serving and streaming
- Quantization tradeoffs · browse: Quantization for inference
Build real AI agents — III — Agents & agentic systems
Browse: Agents and the ReAct loop · LangGraph and LangChain patterns
- LLM vs agent; loops; stop conditions · browse: Agents and the ReAct loop
- Tools (schema, validation, side effects) · browse: Agents and the ReAct loop, Structured outputs
- Planning vs reactive · browse: Agents and the ReAct loop, Multi-agent orchestration
- Memory · browse: Context engineering, Redis for AI caching
- Human-in-the-loop · browse: Guardrails and safety systems
- Trace and debug trajectories · browse: OpenTelemetry for LLMs, Evals fundamentals
- Framework packaging · browse: LangGraph and LangChain patterns
Skills, MCP, context engineering — III — Agents & agentic systems
Browse: Context engineering · Model Context Protocol (MCP)
- Skill vs tool vs prompt · browse: Context engineering
- Author skills · browse: Context engineering
- Dynamic skill loading · browse: Context engineering
- MCP servers and clients · browse: Model Context Protocol (MCP)
- Context budgets · browse: Context engineering
- Coding-agent skill patterns · browse: Model Context Protocol (MCP)
Agentic workflows & multi-agent — III — Agents & agentic systems
Browse: Multi-agent orchestration · RAG building blocks · LangGraph and LangChain patterns
- Workflow patterns · browse: Multi-agent orchestration
- Shared state and handoffs · browse: Multi-agent orchestration
- Agentic RAG · browse: RAG building blocks, Hybrid search and rerankers
- Long-running / checkpoint / resume · browse: Multi-agent orchestration, Kafka for evented AI
- Failure isolation · browse: Multi-agent orchestration, Guardrails and safety systems
- Harness thinking · browse: Multi-agent orchestration, Evals fundamentals
Evals, guardrails, safety — IV — Production + industry
Browse: Evals fundamentals · Guardrails and safety systems
- Offline vs online evals · browse: Evals fundamentals
- LLM-as-judge · browse: Evals fundamentals
- Component vs E2E · browse: Evals fundamentals, RAG building blocks
- Injection / jailbreaks · browse: Guardrails and safety systems
- Guardrails · browse: Guardrails and safety systems
- Canaries · browse: Evals fundamentals
Deploy, cost, latency, observability — IV — Production + industry
Browse: Serving and streaming · Networking for AI apps · Cost and latency routing · OpenTelemetry for LLMs
- API + streaming · browse: Serving and streaming, Networking for AI apps
- Cost accounting · browse: Cost and latency routing
- Caching and latency · browse: Redis for AI caching, Cost and latency routing
- Tracing tool/LLM spans · browse: OpenTelemetry for LLMs
- Retries, rate limits, idempotency · browse: Networking for AI apps
- Runbook · browse: MLflow for LLMOps
How real companies use AI — IV — Production + industry
Browse: vLLM · Model Context Protocol (MCP) · Multi-agent orchestration · Evals fundamentals
- Netflix — in-house LLM serving · browse: vLLM, KV-cache, prefill, and decode
- Uber — agent platform, MCP, gateway · browse: Model Context Protocol (MCP), Guardrails and safety systems
- Meta — LLM-scale training literacy · browse: Scaling laws and compute, Ray
- Stripe — coding agents / harness · browse: Multi-agent orchestration, Evals fundamentals
- Shopify — production merchant agent · browse: Agents and the ReAct loop, Evals fundamentals
- Airbnb — verify-heavy workflows · browse: Multi-agent orchestration
- NVIDIA — agent toolkits and eval/obs · browse: TensorRT-LLM and SGLang, OpenTelemetry for LLMs
- Google — agentic training / RL direction · browse: Alignment basics — RLHF and DPO, Reasoning and test-time compute
AI system design — IV — Production + industry
Browse: LLM project lifecycle · Privacy and data for AI apps · Fine-tuning with LoRA and QLoRA
- Build vs buy · browse: Open-weight models vs closed APIs, LLM project lifecycle
- UX of uncertainty, citations, feedback · browse: RAG building blocks
- Data flywheels · browse: Evals fundamentals, MLflow for LLMOps
- Privacy / retention sketch · browse: Privacy and data for AI apps
- Capstone proposal · browse: LLM project lifecycle
What's coming + staying current — V — Frontier + capstone
Browse: Reasoning and test-time compute · Multimodal basics · Guardrails and safety systems
- Reasoning models / test-time compute · browse: Reasoning and test-time compute
- Multimodal agents · browse: Multimodal basics, Agents and the ReAct loop
- On-device / small models · browse: Quantization for inference, Ollama
- Synthetic data loops · browse: Fine-tuning with LoRA and QLoRA, Evals fundamentals
- Safety, regulation, dual-use · browse: Guardrails and safety systems, Privacy and data for AI apps
- Personal radar — Guided-only: how to stay current
Capstone — V — Frontier + capstone
Browse: LLM project lifecycle · Evals fundamentals · Agents and the ReAct loop
- Capstone requirements and pick-two · browse: LLM project lifecycle
- Architecture and eval plan · browse: Evals fundamentals
- Build and harden · browse: Guardrails and safety systems, Serving and streaming
- Write-up and demo — Guided-only deliverable
Real-world examples (aligned)
- Evals as release gates: shipping LLM changes without vibes · guided: Evals, guardrails, safety · browse: Evals fundamentals, Guardrails and safety systems
- Open-source stack for an AI feature: vLLM, Ollama, and graph orchestrators · guided: Build & serve your SLM · browse: Ollama, vLLM, LangGraph and LangChain patterns
- Multi-tenant AI SaaS: isolation, quotas, and noisy neighbors · guided: Deploy, cost, latency, observability, AI system design · browse: Networking for AI apps, Privacy and data for AI apps
- Fine-tune vs prompt vs RAG: a decision framework · guided: Build & serve your SLM, RAG, Talk to models in the real world · browse: Fine-tuning with LoRA and QLoRA, RAG building blocks
- Agents in production: ReAct loops, timeouts, and human-in-the-loop · guided: Build real AI agents, Agentic workflows & multi-agent, How real companies use AI · browse: Agents and the ReAct loop, Multi-agent orchestration
- Observability for LLM apps: traces, evals, and user feedback · guided: Deploy, cost, latency, observability, Evals, guardrails, safety · browse: OpenTelemetry for LLMs, Evals fundamentals
- Cost control for LLM apps: cache, shrink, and route · guided: Deploy, cost, latency, observability · browse: Cost and latency routing
- Marketplace ranking meets LLMs: Uber/Airbnb-style re-rank patterns · guided: RAG · browse: Hybrid search and rerankers
- Building a ChatGPT-like product: streaming, tools, and memory · guided: Talk to models in the real world, Build real AI agents · browse: Serving and streaming, Agents and the ReAct loop
- Production RAG: chunking, hybrid search, rerank, and eval gates · guided: RAG · browse: Vector databases — what, why, and how, RAG building blocks
- Netflix-style LLM gateway: batching, KV cache, and one API · guided: How real companies use AI, Deploy, cost, latency, observability · browse: vLLM, KV-cache, prefill, and decode