Lessons Map
AI engineering track
AI Engineering: LLM Basics & Prompting Learn the foundations of large language models and prompting. Understand how LLMs work, how to use them effectively, and how to get reliable, high-quality outputs.
Track Progress
0 of 124 lessons completed Continue learning → Start with the first lesson
00 AI Engineer Intro Build an accurate mental model of what an AI engineer actually does before going deeper into how the underlying models work.
1 lessons 01 Foundations: how LLMs work Build an accurate mental model of what a model actually does underneath the API, then use that model to prompt it well.
16 lessons How LLMs Work
Tokens & Tokenization Autoregressive Generation: Training vs Inference Sampling: Temperature, Top-K, Top-P, Repetition Penalty Context Window & Position Handling Core Terminology Map Completion vs Chat vs Reasoning Models Structured Output & Constrained Generation Prompt Anatomy
Prompt Anatomy: System, Context, Constraints, Output Format Prompt Structure, Versioning & Delimiters Prompting Techniques
Zero-Shot & Few-Shot Prompting Chain-of-Thought CoT Variants: Self-Consistency, Tree of Thoughts, Reflexion ReAct: Reasoning + Acting Model Interaction
Function / Tool Calling Prompt Caching Streaming Responses 02 Models & Embeddings Know the model and embedding landscape well enough to choose deliberately instead of defaulting to whatever is easiest to reach for.
14 lessons Model Landscape
Model Types: Pretrained, Closed vs Open, Self-Hosted Closed Models: Claude, GPT/o-series, Gemini, Cohere, Mistral Open Models: Llama, DeepSeek, Qwen, Gemma Platforms: Hugging Face, Ollama, LM Studio, OpenRouter APIs & SDKs: Responses, Messages, Gemini, HF Inference, OpenAI-compatible Choosing a Model
Reading Benchmarks Honestly Choosing a Model Embeddings
What Embeddings Are Similarity Metrics: Cosine, Dot, Euclidean Embedding Models: Proprietary & Open Late Interaction & ColBERT Vector Databases
Vector Databases & Indexes ANN Indexing: HNSW, IVF, Quantization Implementing Vector Search End-to-End 03 Retrieval & RAG Build and tune a retrieval pipeline deliberately, stage by stage, instead of treating RAG as a single black-box feature.
11 lessons What RAG Is, Use Cases, RAG vs Fine-Tuning
What RAG Is, Use Cases, RAG vs Fine-Tuning Implementing RAG
The RAG Pipeline: Chunk, Embed, Index, Retrieve, Generate Chunking Strategies: Fixed, Recursive, Semantic, Structural Hybrid Search: BM25 + Dense, RRF Fusion Reranking: Cross-Encoders, ColBERT, Commercial Rerankers Query Transformation: Rewriting, HyDE, Multi-Query, Decomposition Metadata Filtering & Access-Control-Aware Retrieval Evaluating & Advanced RAG
Retrieval Evaluation: Recall@k, MRR, NDCG, Context Precision/Recall Graph RAG Semantic Caching RAG Frameworks: LangChain, LlamaIndex, Haystack vs Raw SDK
RAG Frameworks: LangChain, LlamaIndex, Haystack vs Raw SDK 04 Agents Know when an agent is actually the right architecture, build the loop by hand before reaching for a framework, and add structure (planning, HITL, multi-agent) only when a task genuinely needs it.
10 lessons Agent Fundamentals
What Agents Are & When to Use Them Tools & Function Calling as Agent Primitive The Agent Loop: Plan, Act, Observe Building Agents
Building Agents From Scratch (No Framework) Planning Loops, Reflection & Self-Correction Error Handling, Retries & Graceful Degradation Advanced Agents
Agentic RAG Human-in-the-Loop & Approval Gates Multi-Agent Systems Orchestration Frameworks: LangGraph, CrewAI, AutoGen, Agent SDKs
Orchestration Frameworks: LangGraph, CrewAI, AutoGen, Agent SDKs 05 Evaluation Build the measurement discipline, observability, metrics, golden sets, judges, regression gates, that turns prompt and model changes from guesswork into an engineering process.
9 lessons Observability
Observability: Tracing, Cost & Latency Monitoring Eval & Observability Tooling: LangSmith, Langfuse, DeepEval, RAGAS Evaluation
Eval Metrics: Deterministic, Model-Based, Human LLM-as-Judge: Calibration, Bias, Failure Modes Building & Maintaining Golden Datasets Testing & Regression
Offline vs Online Eval & A/B Testing Regression Suites for Prompts & Models Testing Nondeterministic Agent Systems Eval-Driven Development as a Workflow
Eval-Driven Development as a Workflow 06 Security Know the attack surface an LLM system actually has, prompt injection, exfiltration, excessive agency, and build the layered defenses, least privilege, sandboxing, compliance controls, red-teaming, that contain it.
10 lessons Attack Surface
OWASP Top 10 for LLM Applications Prompt Injection: Direct, Indirect & Tool-Output Data Exfiltration Paths in Agents Agent Security
Tool-Use Sandboxing & Least-Privilege Agents Secrets & Credential Handling in Agent Systems Multi-Tenant Isolation Compliance & Governance
PII, Data Residency & Governance (SOC 2, GDPR, HIPAA) Content Moderation APIs Testing & Fairness
Adversarial Testing & Red-Teaming Bias & Fairness 07 Context Engineering Treat a model's context window as a designed, bounded, security-aware artifact, not an accumulation of whatever happened to be available.
14 lessons Fundamentals
Context vs Prompt Engineering Context Sources & Retrieval Context Layers & Assembly Memory
Memory Systems in Context Agent Memory Architecture State & Historical Context Techniques
Context Compaction & Summarization Long-Context Processing Multi-Agent Context Sharing Context Isolation Evaluation & Security
Context Failure Modes Context Evaluation Context Security (Indirect Injection) Context Warehouses & Observability Tooling
Context Warehouses & Observability Tooling 08 Model Context Protocol Understand MCP's host/client/server architecture well enough to build a server, wire a client into an agent, and treat every server as a security boundary rather than a trusted dependency.
6 lessons Core Components
MCP Core Components: Host, Client, Server, Data & Transport MCP Transports: stdio, HTTP/SSE, Streamable HTTP Developing with MCP
Building an MCP Server Building an MCP Client Security & Ecosystem
MCP Security Model MCP Ecosystem: Remote Servers & Registries 09 Deployment Take an agent from a local prototype to a monitored, secured, rollback-able production service, through local dev, containerization, hosting, state, queues, secrets, observability, and versioning in order.
9 lessons Environment
Local Development Containerization Where Agents Run: Serverless, Persistent, Managed Runtime Concerns
State & Session Management for Long-Running Agents Queues, Background Jobs & Async Execution Secrets & Credential Management Operations
Logging, Tracing & Alerting From Day One Versioning & Rollback of Prompts, Tools & Agents End-to-End Worked Example: Empty Dir to Deployed Agent
End-to-End Worked Example: Empty Dir to Deployed Agent 10 Enterprise Deployment Scale a deployed agent to enterprise constraints: a defensible reference architecture, private/on-prem delivery, model gateways, cost and throughput discipline, and CI/CD plus monitoring that treats prompts like code.
9 lessons Architecture
Reference Architectures for LLM Systems Private/VPC, On-Prem & Air-Gapped Deployment Model Gateways, Routing, Fallbacks, Multi-Provider Abstraction Cost & Performance
Prompt & Semantic Caching Batching, Streaming & Throughput Cost Modeling, Latency Budgets & Model Selection as Cost Lever Operations & CI/CD
Containerization, Serverless vs Persistent, Autoscaling CI/CD for Prompt & Agent Changes Production Monitoring & Alerting 11 Adaptation & Multimodal Know when to escalate past prompting and RAG into actually adapting a model's weights, run that adaptation end to end, and work with images, video, and speech alongside text.
11 lessons Fine-Tuning
Fine-Tune vs RAG vs Prompt: Decision Framework LoRA & QLoRA Distillation Synthetic Data Generation Structured Output & Constrained Decoding Running a Fine-Tune End-to-End Multimodal
Image Understanding Image Generation Video Understanding Speech: TTS & STT (Whisper) AI Dev Tools: Claude Code, Codex, Cursor
AI Dev Tools: Claude Code, Codex, Cursor 12 Forward-Deployed Craft Scope a vague enterprise ask correctly, translate it into a tractable architecture, communicate the tradeoffs to executives, and demo it honestly on the way to production.
4 lessons Discovery & Scoping
Requirements Discovery & Scoping with Enterprise Stakeholders Translating Vague Asks into Technical Architectures Communication & Demos
Technical Storytelling for Executives Demo Engineering & POC-to-Production Transitions 
Pathweys