Skip to content
All paths
Applied AI Engineer

Build AI systems that hold up in production.

Start free with how language models work, then build the whole Applied AI spine: APIs, tools, retrieval and agents. You write your first eval early, because evals are what separate an Applied AI Engineer from a demo builder. Then come the frontier platforms, production reliability, inference and fine-tuning, the agent engineering AI labs ship today, and system design. It ends with interview prep for AI labs and AI-first product companies.

Start the first lessonSee pricing

79 lessons in 12 milestones, plus 4 optional electives. Starts free; the full path needs the Mastery plan.

By the end you can
  • Ship LLM features on Claude, OpenAI or Gemini with tools, structured output and retries
  • Build retrieval that answers from sources and refuses when it doesn't know
  • Write evals first: golden sets, LLM judges and CI gates that catch regressions
  • Design agents and harnesses with memory, recovery and safe tool permissions
  • Own production: observability, guardrails, cost, latency, inference and fine-tuning
  • Defend a system design and clear an Applied AI interview loop

The roadmap

Work through it in order, or jump to what you need. Open a milestone to see its lessons.

Choose your starting pointyou can change it any time
What do you already know?
  1. Level 1 of 4

    Foundation

    You can explain how a language model works and call one safely.

  2. AI foundations

    Explain how a language model works, call one, and spot when it's making things up.

    Start: What AI Actually Is
    8 lessons · about 5 h 10 min
    1. What AI Actually Is
    2. Inside a Language Model
    3. Tokens, Context & Embeddings
    4. Prompting Basics
    5. Calling a Model
    6. Limits & Hallucination
    7. The Modern AI Toolkit
    8. Evaluating & Choosing Models
  3. Level 2 of 4

    Junior Applied AI Engineer

    You can build tool-using, retrieval-backed features and prove with an eval that they work.

  4. LLM building blocks

    Applied AI plan

    Prompting, APIs, tools, reasoning and embeddings, done properly.

    Start: Software Engineering Practices
    6 lessons · about 2 h 55 min
    1. Software Engineering Practices
    2. Prompt Engineering & Structured Output
    3. LLM APIs & Model Integration
    4. Tool Use & Function Calling
    5. Basic Reasoning Techniques
    6. Vector Embeddings & Semantic Search

  5. Data and retrieval

    Applied AI plan

    RAG pipelines, memory and vector search that return the right context.

    Start: RAG Pipelines
    4 lessons · about 2 h 10 min
    1. RAG Pipelines
    2. Memory Management & Context Engineering
    3. Knowledge Bases & Vector Databases
    4. Data Ingestion & Preprocessing

  6. Your first eval

    Mastery plan

    Build a golden set, score it with metrics and an LLM judge, and know when a change made things worse.

    Start: Evaluation Frameworks & Metrics
    4 lessons
    1. Evaluation Frameworks & MetricsApplied AI
    2. Golden Datasets & LabelingMastery
    3. Metrics & ScoringMastery
    4. LLM-as-Judge & Bias ControlMastery
  7. Level 3 of 4

    Mid Applied AI Engineer

    You can ship agents on any frontier platform and run them in production: reliable, observable, safe and affordable.

  8. Agents and architecture

    Applied AI plan

    Routing, planning, multi-agent systems and MCP.

    Start: Workflow Orchestration & Routing
    5 lessons · about 2 h 30 min
    1. Workflow Orchestration & Routing
    2. Planning, Reflection & Multi-Step Reasoning
    3. Single & Multi-Agent Systems
    4. Model Context Protocol (MCP) & Tool Ecosystems
    5. Parallelization & Resource-Aware Execution

  9. Fluent on every frontier platform

    Mastery plan

    Build on Claude, OpenAI and Gemini, and hide them behind one layer that fails over.

    Start: Multi-Provider Abstraction Layers
    5 lessons · about 4 h
    1. Multi-Provider Abstraction Layers
    2. The Claude Platform
    3. The OpenAI Platform
    4. Google Gemini & Agent Platform
    5. MCP Apps for Customers

  10. Production and reliability

    Mastery plan

    Recovery, observability, guardrails, cost, agent evals, CI gates and prompt-injection defence.

    Start: Exception Handling, Recovery & Human-in-the-Loop
    11 lessons
    1. Exception Handling, Recovery & Human-in-the-LoopApplied AI
    2. Observability, Monitoring & LoggingApplied AI
    3. Guardrails, Safety & AlignmentApplied AI
    4. Cost Optimization & Latency EngineeringApplied AI
    5. RAG & Agent EvaluationMastery
    6. Offline vs Online EvalMastery
    7. Statistical Power & SignificanceMastery
    8. CI Eval GatesMastery
    9. Drift DetectionMastery
    10. Prompt Injection & the LLM Attack SurfaceSystems
    11. Sandboxing Untrusted Code & ToolsSystems

  11. Inference, fine-tuning and scale

    Mastery plan

    Deploy and serve models, tune inference, version AI changes and decide when to fine-tune.

    Start: Deployment & Infrastructure
    10 lessons · 4 electives
    1. Deployment & InfrastructureApplied AI
    2. Inference OptimizationApplied AI
    3. CI/CD, Testing & Versioning for AIApplied AI
    4. Fine-tuning Strategies vs Prompt/RAGApplied AI
    5. Prefill vs DecodeSystems
    6. The KV CacheSystems
    7. Batching & Continuous BatchingSystems
    8. QuantizationSystems
    9. Throughput, Latency & $/Token EconomicsSystems
    10. GPU Serving (vLLM / TGI)Mastery
    11. Electives: optional, they don't count toward this milestone
      GPU Architecture & the RooflineSystems
    12. VRAM & the GPU Memory ModelSystems
    13. Attention Cost & Context ScalingSystems
    14. Multi-GPU ParallelismSystems
  12. Level 4 of 4

    Senior Applied AI Engineer

    You can build frontier agent systems, design and defend an AI architecture, and clear the interview loop.

  13. Frontier agent engineering

    Mastery plan

    Build what AI labs ship today: long-running agent harnesses, post-trained models, voice agents and computer-use agents.

    Start: Agent Harness Engineering
    4 lessons
    1. Agent Harness Engineering
    2. Post-Training in Practice
    3. Realtime & Voice Agents
    4. Computer-Use & Browser Agents

  14. AI system design

    Mastery plan

    Turn a vague product ask into an architecture with latency, cost, eval and failure budgets, and defend it.

    Start: AI System Design & Trade-off Analysis
    9 lessons
    1. AI System Design & Trade-off AnalysisApplied AI
    2. Requirements & ConstraintsMastery
    3. Retrieval Architecture at ScaleMastery
    4. Latency Budgets & the Critical PathMastery
    5. Multi-Agent Orchestration DesignMastery
    6. Eval & Safety PlanningMastery
    7. Reliability, Degradation & Failure DesignMastery
    8. Cost Modeling at 10×Mastery
    9. The Design ReviewMastery

  15. Advanced systems and leadership

    Mastery plan

    Multimodal, adaptive and experimental systems, AI product judgement, and leading the work.

    Start: Prioritization, Goal Setting & Monitoring
    8 lessons
    1. Prioritization, Goal Setting & MonitoringApplied AI
    2. Multimodal & Emerging CapabilitiesApplied AI
    3. Learning & Adaptation MechanismsApplied AI
    4. Exploration, Experimentation & InnovationApplied AI
    5. Mentorship, Documentation & CollaborationApplied AI
    6. UX for AI SystemsMastery
    7. Communicating UncertaintyMastery
    8. Feedback, Correction & the Data FlywheelMastery

  16. Get the offer

    Mastery plan

    Practise the loop: the LLM coding round, LLM system design, the eval and debugging round, and your project deep-dive.

    Start: The Applied AI Interview Loop
    5 lessons
    1. The Applied AI Interview Loop
    2. The LLM Coding Round
    3. LLM System Design Round
    4. The Eval & Debugging Round
    5. Project Deep-Dive & Behavioral

  17. Final project

    Ship a real system on real infrastructure. Each project has an architecture diagram and a step-by-step build guide.

    Production RAG Over Ugly Real DocumentsA law / insurance / clinical firm hands you 500 genuinely ugly PDFs, scans, multi-column, tables, and demands grounded answers with citations and zero hallucination tolerance.A Multi-Agent System on a Real FrameworkAutomate a real knowledge-work task end-to-end: a competitive-research report, a codebase migration, or a due-diligence memo.Evals-as-a-Discipline: a CI Gate for an LLM FeatureYour team ships prompt and model changes straight to prod with no safety net. Build the gate that stops a regression.Post-Train a Small Model and Prove ItA narrow, high-volume task where a fine-tuned small model might beat prompting a frontier model on cost and quality.

Start where you are.

Your progress is saved as you go, and every lesson you finish counts toward every path that includes it.

Compare paths