AI Engineering learning roadmap.
Build from LLM fundamentals to RAG, LangGraph agents, secure workflows, and production AI systems. Work through 32 modules, then use the connected questions to check what you can explain without the notes.
- STEP 1
Start here
- How to use this AI Engineering guide · 2 min
- Priority key · 2 min
- Contents · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 2
AI and LLM fundamentals
- AI, machine learning, deep learning, and generative AI · 2 min
- LLMs and how they work · 2 min
- Tokens and token usage · 2 min
- Context window and context construction · 2 min
- Temperature and top-p · 2 min
- System, user, assistant, and tool messages · 2 min
- Prompt engineering, zero-shot, and few-shot prompting · 2 min
- Structured output and JSON output · 3 min
- Function calling and tool calling · 2 min
- Streaming · 2 min
- Multimodal models, vision, and audio · 2 min
- Model selection, routing, fallback, latency, and cost · 2 min
- Provider ecosystems: OpenAI, Anthropic, Gemini, and Azure OpenAI · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 3
Python for AI engineers
- Python basics, lists, dictionaries, and sets · 3 min
- Functions, classes, modules, and exceptions · 2 min
- Virtual environments, pip, and project structure · 2 min
- Async/await, type hints, and HTTP clients · 2 min
- Pydantic and FastAPI · 2 min
- Basic NumPy, Pandas, and Jupyter · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 4
Embeddings
- Embeddings, vectors, dimensions, and semantic similarity · 2 min
- Similarity and cosine similarity · 2 min
- Embedding models, generation, storage, and limitations · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 5
Vector databases and search
- Vector databases, vector search, and semantic search · 2 min
- Keyword search and BM25 · 2 min
- Hybrid search and metadata filtering · 2 min
- Reranking · 2 min
- Choosing a vector/search store · 3 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 6
RAG, beginner to advanced
- RAG: three levels of understanding · 3 min
- Document ingestion, PDF processing, and parsing · 2 min
- Chunking, chunk size, overlap, and metadata · 2 min
- Retrieval, context construction, and prompt construction · 2 min
- Citations and source attribution · 2 min
- Query rewriting, multi-query retrieval, and query decomposition · 2 min
- Hybrid RAG, reranking, and parent-child retrieval · 2 min
- Multi-hop RAG and agentic RAG · 2 min
- Corrective RAG, self-reflective RAG, and context compression · 2 min
- Graph RAG · 2 min
- RAG evaluation and comparison with alternatives · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 7
Agents and agent patterns
- Agents, chatbots, and workflows · 2 min
- Agent loop, reasoning/action/observation, and ReAct · 2 min
- Tools, planning, state, memory, and autonomous actions · 2 min
- Human-in-the-loop · 2 min
- Router pattern · 2 min
- Planner/executor pattern · 2 min
- Supervisor and handoff patterns · 2 min
- Reflection and critic patterns · 2 min
- Multi-agent and hierarchical agents · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 8
LangChain
- LangChain overview, models, prompts, and chains · 2 min
- LangChain tools, retrievers, retrieval, and structured output · 2 min
- LangChain agents versus simple code · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 9
LangGraph, in depth
- What LangGraph is and why it exists · 3 min
- State and reducers · 2 min
- Nodes, edges, conditional edges, and routing · 2 min
- Practical TypeScript graph with an approval pause · 2 min
- Checkpoints, persistence, and memory · 2 min
- Interrupts and human review · 2 min
- Tool integration and the agent loop in a graph · 2 min
- Subgraphs and multi-agent graphs · 2 min
- Streaming graph events · 2 min
- Errors, retry logic, and recovery · 2 min
- Production architecture and choosing LangGraph · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 10
Modern agent frameworks
- OpenAI Agents SDK · 2 min
- Microsoft Agent Framework: agents and workflows · 2 min
- Semantic Kernel and AutoGen · 2 min
- CrewAI and Google ADK · 3 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 11
Deterministic and agentic workflows
- Deterministic workflows · 2 min
- Agentic workflows · 2 min
- Hybrid workflows and state machines · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 12
Workflow orchestration
- State machines, DAGs, and event-driven workflows · 2 min
- Long-running workflows, retries, timeouts, and schedules · 2 min
- Idempotency, compensation, and human approvals · 2 min
- Choosing orchestration tools · 3 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 13
Model Context Protocol (MCP)
- MCP architecture: host, client, and server · 2 min
- MCP tools and tool discovery · 2 min
- MCP resources and prompts · 2 min
- Authentication, authorization, and permissions · 2 min
- MCP versus REST, function calling, and plugins · 2 min
- Conceptual MCP server: four business tools · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 14
AI memory
- Context, short-term memory, and conversation memory · 2 min
- Long-term, semantic, episodic, and user-profile memory · 2 min
- State persistence and checkpoints · 2 min
- Context summarization and memory retrieval · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 15
AI security
- Prompt injection, indirect injection, and jailbreaking · 2 min
- Tool poisoning and excessive agency · 2 min
- Data leakage, PII, sensitive data, and secrets · 2 min
- Authentication, authorization, least privilege, and tenant isolation · 2 min
- Validation, rate limiting, and secure database access · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 16
Guardrails
- Input, output, and tool guardrails · 2 min
- Schema validation: Zod and Pydantic · 2 min
- Moderation and PII detection · 2 min
- Hallucination mitigation, business rules, and approval · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 17
AI evaluation
- Golden datasets, test cases, and regression testing · 2 min
- Retrieval precision, recall, relevance, and ranking · 2 min
- Faithfulness, answer correctness, and citation quality · 2 min
- Tool-call accuracy and task completion · 2 min
- LLM-as-a-judge and human evaluation · 2 min
- Evaluation tools: LangSmith, Langfuse, and Arize Phoenix · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 18
AI observability
- Traces, agent runs, LLM/tool calls, and state transitions · 2 min
- Latency, tokens, cost, errors, and service objectives · 2 min
- Langfuse, LangSmith, OpenTelemetry, and Sentry · 2 min
- A realistic production trace · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 19
Cost and performance
- Token optimization, prompt compression, and context trimming · 2 min
- Small versus large models and practical routing · 2 min
- Exact caching, prompt caching, and semantic caching · 2 min
- Batch processing, concurrency, rate limits, streaming, and fallback · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 20
AI gateways
- Gateways, routing, fallback, quotas, and observability · 2 min
- LiteLLM, Portkey, and cloud AI gateways · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 21
Multimodal AI
- Vision and image understanding · 2 min
- PDFs, OCR, and document AI · 2 min
- Speech-to-text, text-to-speech, and audio understanding · 2 min
- Voice agents and video understanding · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 22
Local AI
- Local LLMs, API models, and self-hosting · 2 min
- Ollama, Hugging Face, and Transformers · 2 min
- vLLM and GPU serving · 2 min
- Quantization, GGUF, and local embeddings · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 23
Fine-tuning
- Fine-tuning and when it is justified · 2 min
- LoRA, QLoRA, and PEFT · 2 min
- Dataset preparation, evaluation, and RAG comparison · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 24
Cloud AI
- Azure OpenAI and Azure AI Foundry / Microsoft Foundry · 2 min
- Azure AI Search, Functions, Container Apps, Key Vault, and Entra ID · 2 min
- AWS Bedrock, Lambda, ECS, S3, OpenSearch, and IAM · 2 min
- Google Cloud: Vertex AI, Cloud Run, and Cloud Functions · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 25
MERN and AI
- Shared MERN architecture · 2 min
- AI chatbot · 2 min
- Streaming AI chatbot · 2 min
- RAG chatbot · 2 min
- AI search · 2 min
- AI summarizer · 2 min
- AI extractor · 2 min
- AI classifier · 2 min
- Customer-support agent · 2 min
- CRM assistant · 2 min
- Email assistant · 2 min
- Workflow automation and introducing Python/FastAPI · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 26
MongoDB and AI
- Atlas, document storage, metadata, and vector search · 2 min
- Conversations, agent state, and user memory · 2 min
- Multi-tenancy and comparison with PostgreSQL + pgvector · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 27
AI system design
- Scalability, availability, reliability, and latency · 2 min
- Queues, caching, databases, and vector indexes · 2 min
- Providers, failover, security, multi-tenancy, and observability · 2 min
- Design 1: RAG chatbot · 2 min
- Design 2: Customer-support agent · 2 min
- Design 3: Insurance claim agent · 2 min
- Design 4: AI CRM agent · 2 min
- Design 5: Document processing system · 2 min
- Design 6: Email agent · 2 min
- Design 7: Research agent · 2 min
- Design 8: Multi-agent system · 2 min
- Design 9: Enterprise knowledge assistant · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 28
Insurance AI reference project
- The overall insurance assistant · 3 min
- Which parts are deterministic and which are agentic? · 2 min
- Tools and their safe contracts · 2 min
- Data model · 2 min
- API design · 2 min
- Two complete request walkthroughs · 2 min
- Failure handling, evaluation, and operational review · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 29
Practical project roadmap
- Project 1 — Basic LLM chatbot · 2 min
- Project 2 — Streaming AI chatbot · 2 min
- Project 3 — PDF RAG assistant · 2 min
- Project 4 — Production RAG application · 2 min
- Project 5 — Tool-calling agent · 2 min
- Project 6 — LangGraph agent · 2 min
- Project 7 — Multi-agent customer support · 2 min
- Project 8 — MCP server plus AI agent · 2 min
- Project 9 — Deterministic plus agentic workflow · 2 min
- Project 10 — Production insurance AI platform · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 30
Interview preparation
- LLMs · 2 min
- Prompt engineering · 2 min
- RAG · 2 min
- Embeddings · 2 min
- Vector databases · 2 min
- LangChain · 2 min
- LangGraph · 2 min
- AI agents · 2 min
- MCP · 2 min
- Microsoft Agent Framework · 2 min
- Workflows · 2 min
- Memory · 2 min
- AI security · 2 min
- Guardrails · 2 min
- Evaluation · 2 min
- Observability · 2 min
- Cost optimization · 2 min
- Python · 2 min
- FastAPI · 2 min
- MERN and AI · 2 min
- AI system design · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 31
Senior engineer decision framework
- The decision mindset · 3 min
- Do I need AI? Can deterministic code solve it? · 2 min
- Do I need an LLM? · 2 min
- Do I need RAG? · 2 min
- Do I need tools? Do I need MCP? · 2 min
- Do I need an agent? Do I need LangGraph? · 2 min
- Should this be a workflow? · 2 min
- Where is human approval required? · 2 min
- How do I secure it? · 2 min
- How do I evaluate it? · 2 min
- How do I observe it? · 2 min
- How much does it cost? · 2 min
- How does it scale? · 2 min
- Real client decisions · 2 min
- Stable concepts, framework concepts, and changing APIs · 2 min
- A concrete architecture proposal template · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
- STEP 32
Interview revision cheat sheet
- Core vocabulary · 2 min
- RAG revision · 2 min
- Agents and frameworks · 2 min
- Workflows, memory, and safe operations · 2 min
- Security and guardrails · 2 min
- Evaluation and observability · 2 min
- Cost, multimodal, and deployment · 2 min
- Two-minute insurance system answer · 2 min
- Questions to ask before choosing architecture · 2 min
Checkpoint: explain the ideas above with an example and a limitation before continuing.
Check your understanding
After the lessons, rehearse the 299 connected questions. Revisit the lesson when you cannot yet explain why the answer works.
Practice AI Engineering Review the interview guide