Learning Path
Learn production AI engineering, track by track.
54 concise, production-focused lessons across 6 tracks. Read in order or jump to what you need.
01
AI Engineering Foundations
The core skills every production AI engineer needs — from LLM basics to shipping reliable systems.
9 lessons
02
Building RAG Systems
Design retrieval-augmented generation from chunking and embeddings to reranking and evaluation.
9 lessons
03
Agentic AI & Orchestration
Build agents that plan, use tools, keep memory, and coordinate — without going off the rails.
9 lessons
04
LLMOps & Production
Deploy, monitor, evaluate, and continuously improve LLM products in production.
9 lessons
05
Evaluation, Safety & Guardrails
Measure quality, catch regressions, and add guardrails that hold up under real usage.
9 lessons
06
AI System Design & Interviews
Architecture tradeoffs and interview-ready reasoning for AI-native systems.
9 lessons
AI Engineering Foundations
- What Is AI Engineering?How AI engineering differs from ML research and traditional software, and what the job actually involves.6 min
- Context Windows and TokensWhat tokens are, why the context window matters, and how to budget it deliberately.7 min
- Prompting FundamentalsCore prompting patterns every AI engineer needs — system prompts, few-shot examples, chain-of-thought, and role framing.8 min
- Choosing the Right ModelA decision framework for selecting among frontier, mid-tier, and small models based on task requirements.7 min
- Calling the LLM APIHow to make robust API calls to LLM providers — authentication, retries, streaming, and error handling.9 min
- Structured Outputs and JSON ModeMake LLMs return machine-readable data reliably using JSON mode, function calling, and output validation.8 min
- Cost and Latency FundamentalsHow LLM pricing works, where latency comes from, and engineering practices that control both without sacrificing quality.8 min
- Token Cost Estimation Before You BuildBuild a reliable cost model for your LLM feature before launch so you can set pricing, budget, and optimization priorities.7 min
- Streaming Responses and UXImplement token streaming end to end so users see incremental output, and design the UI patterns that make streaming feel polished.8 min
Building RAG Systems
- RAG Overview and When to Use ItWhat retrieval-augmented generation is, the problems it solves, and when it is the right architectural choice.7 min
- Chunking StrategiesHow to split documents into chunks that retrieve well — fixed-size, sentence-boundary, semantic, and hierarchical approaches.9 min
- Embeddings ExplainedWhat text embeddings are, how they encode meaning, and how to choose and call an embedding model for RAG.8 min
- Vector Search BasicsHow vector databases find nearest neighbors, key indexing algorithms, and practical guidance for choosing a vector store.9 min
- Reranking for PrecisionHow cross-encoder rerankers improve retrieval precision, where they fit in the RAG pipeline, and when to use them.8 min
- Evaluating RAG QualityMetrics and evaluation frameworks for measuring retrieval accuracy, answer faithfulness, and end-to-end RAG pipeline quality.10 min
- Common RAG Failure Modes and FixesA practical guide to the most frequent ways RAG systems break in production and how to diagnose and fix each one.10 min
- Hybrid Search: BM25 and Dense RetrievalCombine keyword-based BM25 and embedding-based dense search to improve recall on queries where either method alone falls short.9 min
- Measuring Retrieval QualityApply Recall@K, MRR, NDCG, and context precision to quantify retrieval performance and drive systematic improvement.8 min
Agentic AI & Orchestration
- The Agent LoopUnderstand the perceive-think-act cycle that drives every LLM agent and why it differs from a single model call.7 min
- Tool Use and Function CallingHow to expose tools to a language model so it can take real-world actions — from search to code execution.9 min
- Planning and Reasoning in AgentsHow ReAct, chain-of-thought, and structured planning patterns help agents decompose and solve complex tasks.8 min
- Memory and Context ManagementTechniques for giving agents persistent, relevant memory across long conversations and multi-session tasks.9 min
- Multi-Agent OrchestrationDesign patterns for coordinating multiple specialized agents — orchestrators, subagents, and parallel execution.10 min
- Guardrails for AgentsPractical techniques to constrain agent behavior, prevent harmful actions, and keep humans in the loop when it matters.8 min
- Evaluating AgentsHow to measure agent quality — trajectory-based evaluation, task completion rates, and building reliable eval harnesses.10 min
- Designing Tool Schemas That Models Use WellWrite tool definitions that make models select, invoke, and interpret tools reliably — avoiding the most common schema design mistakes.8 min
- Agent Observability and TracingInstrument autonomous agents so you can inspect, debug, and monitor every decision, tool call, and reasoning step in production.9 min
LLMOps & Production
- What Is LLMOps?How LLMOps differs from traditional MLOps and what the production lifecycle of an LLM application actually involves.6 min
- Prompt Versioning and ManagementTreat prompts as code: how to version, review, test, and deploy prompt changes with the same rigor as software releases.8 min
- Observability and TracingInstrument your LLM application so you can see exactly what happened in any request — prompts, completions, tool calls, and errors.9 min
- Cost and Latency MonitoringSet up dashboards and alerts to track token spend and response latency before they become production surprises.8 min
- Canary Releases and A/B TestingRoll out prompt and model changes safely using canary traffic splitting, A/B experiments, and shadow mode evaluation.8 min
- Caching Strategies for LLM SystemsReduce latency and cost with prompt caching, semantic caching, and result caching — and know which to apply where.7 min
- Semantic Caching in DepthBuild a production semantic cache that serves similar LLM queries from prior results, cutting cost and latency for read-heavy workloads.9 min
- Incident Runbooks for LLM OutagesPrepare your team to respond quickly and calmly when an LLM provider goes down or starts returning degraded results.8 min
- Feedback Loops and Dataset CurationClose the production feedback loop by capturing user signals, curating datasets from real traffic, and continuously improving your system.9 min
Evaluation, Safety & Guardrails
- Why Evaluations MatterUnderstand why systematic evaluation is the foundation of trustworthy AI product development.6 min
- Building a Golden DatasetHow to collect, curate, and version a high-quality set of ground-truth examples for evaluation.8 min
- LLM-as-Judge EvaluationUse a capable model as an automated scorer to evaluate outputs too open-ended for deterministic checks.9 min
- Faithfulness and Hallucination DetectionTechniques to detect and measure when a model generates claims not supported by its source context.8 min
- Regression Testing for PromptsHow to catch quality regressions when you change a prompt, swap a model, or update retrieval logic.7 min
- Adding Guardrails to AI SystemsDesign and implement input and output guardrails that enforce safety, policy, and quality constraints.9 min
- Red Teaming AI SystemsStructured adversarial testing to find safety and policy failures before they reach production users.10 min
- Building a Production Eval HarnessDesign and implement an automated evaluation harness that runs on CI, scores outputs consistently, and gates deployments on quality.10 min
- Layered Guardrails in PracticeImplement a defense-in-depth guardrail stack that balances safety, latency, and false-positive rate for production AI systems.9 min
AI System Design & Interviews
- Designing an AI Feature End to EndA practical walkthrough of the decisions required to design and ship a production AI feature from scratch.10 min
- RAG System DesignArchitectural decisions for building retrieval-augmented generation pipelines that are accurate, fast, and maintainable.11 min
- Retrieve vs. Fine-Tune: Making the DecisionA framework for deciding when retrieval-augmented generation is the right choice versus fine-tuning the model.8 min
- Scaling and Latency in AI SystemsTechniques for meeting latency budgets and handling traffic spikes in production AI applications.9 min
- Reliability and Fallback DesignDesign patterns for keeping AI features working gracefully when models fail, timeout, or return unusable outputs.8 min
- An Interview Answer Framework for AI DesignA structured approach to answering AI system design interview questions clearly and completely in 30 to 45 minutes.7 min
- Worked Interview Example: AI Customer SupportA complete walkthrough of designing an AI-powered customer support system from requirements to architecture.12 min
- Multi-Tenant RAG IsolationDesign a RAG system that serves multiple customers from shared infrastructure while guaranteeing that no tenant can access another tenant's data.10 min
- Cost and Latency Capacity PlanningPlan for scale before you need it: how to model token costs, latency budgets, and provider rate limits for growing traffic.9 min