AI Engineering Roadmap

Learn AI Engineering step by step — 62 free lessons with spaced repetition and active recall on RetainHQ, so what you study actually sticks.

Lessons

• Agent autonomy spectrum
• Agent checkpointing & resumability
• Building Eval Harnesses for Multi-Step Tasks
• A Taxonomy of Production Agent Failures
• Sandboxing and Least Privilege for Agents
• Memory as an Actively-Invoked Tool
• Context Bloat: Pruning and Compacting Agent State
• Sandboxing Agent Tool Execution
• Normalizing Agent Traces Before Scoring
• Agent Evals: Trajectory vs Outcome Scoring
• ANN Indexes (HNSW & IVF)
• Caching & Cost Control
• Chain-of-Thought
• Chat Completions API
• Chunking Strategies
• Completion vs Chat Models
• Context Injection & Prompt Assembly
• Context Window
• Cost & Latency Budgeting
• Embed & Retrieve Top-k
• Embeddings
• Evaluation & Test Sets
• Function / Tool Calling
• Grounding with Retrieval
• Guardrails & Validation
• Hallucination Mitigation
• Human-in-the-loop interrupts
• Security: Indirect Prompt Injection via Tool Results
• MCP Architecture: Hosts, Clients, and Servers
• Security: The Confused Deputy Problem in MCP Proxies
• Building Servers: The FastMCP Framework
• The M×N integration problem and protocol necessity
• Observability: Inspecting MCP Traffic
• Agentic Reversals: MCP Sampling and Elicitation
• Security: MCP Tool Poisoning and Schema Attacks
• MCP Primitives: Tools, Resources, and Prompts
• MCP Transports: stdio and Streamable HTTP
• When to Use MCP vs Raw Function Calling
• Agent Memory: Short vs. Long Term
• Metadata Filtering
• Multi-agent orchestration patterns
• Multi-Step Planning
• Observability & Tracing
• Output Formatting & Delimiters
• pgvector, Pinecone, & Chroma
• Prompt Injection
• RAG Evaluation
• Rate Limits & Retries
• ReAct (Reason + Act)
• Reranking
• Preventing Agent Runaway: Budgets and Caps
• Similarity Metrics
• State machines for agents
• Streaming Responses
• Structured Output (JSON Mode)
• System Prompts & Guardrails
• System / User / Assistant Roles
• Temperature & Top-p
• Tokens & tokenization
• Tool Use & The Call Loop
• When multi-agent underperforms
• Zero-shot vs Few-shot