Skip to content
AI ENGINEERINGROADMAP96 lessons
Sign In→
Sign In

The Expedition

Climb the AI engineering stack

11 milestones, sequenced from fundamentals to production systems. Follow the rail or jump to any station.

  1. 01
    🧠Milestone 1Intermediate

    Phase 1 · Foundations & Model Inference

    Understand what an AI engineer ships, then run real inference on pre-trained models — and reason about their token and cost behavior

    • 01The AI Engineer's Roadmap: Skills, Tools & Career Path
    • 02Transformer Architecture: Attention, Positional Encoding & Scale
    • 03Tokenization: BPE, SentencePiece & Vocabulary Design
    • +4 more lessons
    7 lessons · 192mEnter station→
  2. 02
    💬Milestone 2Intermediate

    Phase 2 · Prompting & Structured Output

    Make model output reliable — prompts, tool/function calling, and schema-valid JSON

    • 01Prompt Engineering Fundamentals: Principles & Patterns
    • 02Few-Shot & Chain-of-Thought Prompting
    • 03System Prompt Design: Instructions, Personas & Guardrails
    • +7 more lessons
    10 lessons · 264mEnter station→
  3. 03
    🔍Milestone 3Intermediate

    Phase 3 · Embeddings & RAG

    Ground answers in your own data — embeddings and retrieval with a vector database and a metadata store

    • 01Embeddings
    • 02Embedding Models: Representations, Similarity & Fine-tuning
    • 03Vector Databases: Indexing, ANN Search & Production Patterns
    • +6 more lessons
    9 lessons · 211mEnter station→
  4. 04
    🤖Milestone 4Intermediate

    Phase 4 · Agents & Orchestration

    Build agents that reason and act — tool-use loops, multi-agent coordination, and orchestration

    • 01Agent Architectures: ReAct, Plan-and-Execute & Cognitive Frameworks
    • 02Multi-Agent Systems: Orchestration, Delegation & Communication
    • 03Agent Orchestration: Routing, Handoffs & Supervisor Patterns
    • +4 more lessons
    7 lessons · 265mEnter station→
  5. 05
    🧠Milestone 5Intermediate

    Phase 5 · Long-Term Memory

    Give agents durable recall — semantic, episodic and procedural memory, context-window engineering, and long-term memory stores that survive across sessions

    • 01Agent Memory: Short-term, Long-term & Episodic Memory Systems
    • 02Memory: Persistence as an Agentic Pattern
    • 03Context Engineering: Designing What LLMs See
    • +5 more lessons
    8 lessons · 233mEnter station→
  6. 06
    ⛓Milestone 6Intermediate

    Phase 6 · LangChain & LangGraph

    Build production agent graphs with LangChain and LangGraph — LCEL chains, stateful graphs, D1-checkpointed memory, human-in-the-loop, and tracing

    • 01LangChain Fundamentals: Runnables, LCEL & Composable Chains on the Edge
    • 02LangChain Tools & Retrievers: Grounding Chains in Your Own Data
    • 03LangGraph: Stateful Multi-Agent Graphs for Production AI
    • +6 more lessons
    9 lessons · 107mEnter station→
  7. 07
    🛡Milestone 7Intermediate

    Phase 7 · Evals, Safety & Observability

    Prove it works and ships safely — evals, red-teaming, guardrails, and gateway-level observability

    • 01LLM Evaluation Fundamentals: Metrics, Datasets & Methodology
    • 02Benchmark Design: Contamination, Saturation & Domain-Specific Evals
    • 03LLM-as-Judge: Automated Evaluation, Calibration & Bias
    • +14 more lessons
    17 lessons · 476mEnter station→
  8. 08
    🚀Milestone 8Intermediate

    Phase 8 · Ship to Production

    Take it to production — deploy, scale, cost-control, and CI/CD

    • 01Edge Deployment: On-Device Models, ONNX & WebLLM
    • 02Cost Optimization: Token Economics, Caching & Model Selection
    • 03Scaling & Load Balancing: GPU Clusters, Model Parallelism & Routing
    • +8 more lessons
    11 lessons · 318mEnter station→
  9. 09
    🔧Milestone 9Intermediate

    Appendix · Fine-tuning & Training

    Optional track — customize models with LoRA, RLHF and dataset curation when prompting is not enough

    • 01Fine-tuning Fundamentals: Full, Freeze & Transfer Learning
    • 02LoRA, QLoRA & Adapter Methods: Parameter-Efficient Fine-tuning
    • 03RLHF & Preference Optimization: DPO, ORPO & PPO
    • +3 more lessons
    6 lessons · 166mEnter station→
  10. 10
    ☁Milestone 10Intermediate

    Appendix · Other Clouds & Platforms

    Optional track — GCP, Docker and Kubernetes for portability and comparison off the primary path

    • 01Google Cloud Platform
    • 02Docker
    • 03Kubernetes
    3 lessons · 33mEnter station→
  11. 11
    🏗Milestone 11Intermediate

    Appendix · Engineering & Communication

    Optional track — software-engineering foundations (SOLID, ACID, SQL, CI/CD) plus LlamaIndex and communication

    • 01Microservices
    • 02CI/CD
    • 03Node.js
    • +6 more lessons
    9 lessons · 112mEnter station→
The roadmap

Become an AI engineer,
one lesson at a time.

Start the path
98Lessons
11Skill areas
475K+Words written
AI Engineering

A deep-dive learning path for junior AI engineers — crafted by Vadim Nicolai.

Built with Next.js & Radix UI

Skill areas

  • 🧠Phase 1 · Foundations & Model Inference
  • 💬Phase 2 · Prompting & Structured Output
  • 🔍Phase 3 · Embeddings & RAG
  • 🤖Phase 4 · Agents & Orchestration
  • 🧠Phase 5 · Long-Term Memory
  • ⛓Phase 6 · LangChain & LangGraph
  • 🛡Phase 7 · Evals, Safety & Observability
  • 🚀Phase 8 · Ship to Production
  • 🔧Appendix · Fine-tuning & Training
  • ☁Appendix · Other Clouds & Platforms
  • 🏗Appendix · Engineering & Communication
© 2026 AI Engineering. All rights reserved.Always learning