AiFDE School: Where AI gets real.

What you'll learn

From your first model call to production.

Both courses are synchronous and project-based: everyone logs in at the same time, cameras on, never prerecorded. You build real systems in class, with your instructor, and every project is measured, hardened and demoed.

Each course
54 hours
Classes
22 sessions + a 10-hour hackathon
Structure
6 modules, 3 projects
Hackathon prizes
$10,000 per course

01 · The curriculum

Pick a course. See every session.

You build one healthcare front-office system, session by session, the way forward deployed engineers work: Discovery, Goal, Prototype, Productionize.

Weekend cohort from Sat, Dec 5 · Weekday cohort from Mon, Dec 14 · Course page, schedule and price

What you build, module by module

One system that grows every session.

  1. Module 1 · Sessions 1–4

    The FDE role and how LLMs work

    Route incoming support tickets with versioned, tested prompts, on a model you picked with a bake-off.

  2. Module 2 · Sessions 5–8

    Structured output, context and retrieval

    Pull structured data out of messages, call tools, remember the conversation and answer from clinic documents with citations.

  3. Module 3 · Sessions 9–12

    RAG evaluation and AI agents

    Project 1: a RAG knowledge assistant measured against an evaluation set. Then a front-desk agent that asks staff before changing anything, hardened with guardrails, retries and fallbacks.

    Project 1 demosMini project demos
  4. Module 4 · Sessions 13–16

    MCP and the email agent

    Scheduling and knowledge servers on MCP, attacked, defended and deployed remotely. Project 2: an email agent that routes patient email and drafts replies for staff approval.

    Project 2 demos
  5. Module 5 · Sessions 17–20

    Multi-agent systems, observability and evaluation

    A supervisor agent that calls a billing agent over A2A, a deep agent with memory and planning, tracing with dashboards and alerts, and agent evaluations.

  6. Module 6 · Sessions 21–22

    Production evaluation, security and cost

    Production-ready: an online evaluator, a CI quality gate that blocks bad changes, a red-team test suite and a cost and latency benchmark.

Sun, Jan 24, 2027 · 8 AM–6 PM ET

Capstone Hackathon

Both cohorts together. Design, build, evaluate and present an end-to-end FDE solution in one 10-hour day.

1st$5,000
2nd$3,000
3rd$2,000

Session by session

22 sessions. Six modules.

Show dates for
Module 1The FDE role and how LLMs workDec 5–6
  1. 01Sat Dec 5

    The FDE role and GenAI vs traditional programming

    What an FDE is and why FDEs matter; stages of FDE work; FDE vs other roles; when not to use AI

    LabsSupport-ticket router

  2. 02Sat Dec 5

    How LLMs work

    Tokens and token inflation; embeddings; transformers; mixture of experts; pre-training and post-training

    LabsToken explorer; embeddings

  3. 03Sun Dec 6

    Model selection and inferencing

    Model landscape; open vs commercial models; benchmarks; inference

    LabsModel bake-off

  4. 04Sun Dec 6

    Prompt engineering and prompts as code

    Anatomy of a prompt; few-shot; chain of thought vs reasoning models; prompt caching

    LabsFew-shot router; versioned prompts with evaluation

Module 2Structured output, context and retrievalDec 12–13
  1. 05Sat Dec 12

    Structured output and tool calling

    Why free text breaks; structured output; constrained decoding; tool calling

    LabsData extraction; tool-calling round trip

  2. 06Sat Dec 12

    Context engineering and memory

    Context as a budget; write, select, compress, isolate; short- and long-term memory

    LabsContext budget meter; chat with memory

  3. 07Sun Dec 13

    RAG I: retrieval design and RAG types

    Parsing and chunking; embeddings and vector stores; RAG types

    LabsChunking strategies; index and ask

  4. 08Sun Dec 13

    RAG II: advanced retrieval and grounded generation

    Why naive RAG fails; hybrid search and reranking; grounded generation

    LabsRetriever comparison; grounded, cited answers

Module 3RAG evaluation and AI agentsDec 19–20
  1. 09Sat Dec 19

    RAG III: RAG evaluationProject 1 demos

    What to measure; building an evaluation set

    LabsNaive vs grounded RAG

  2. 10Sat Dec 19

    AI agents I: foundations

    What an agent is; workflow patterns; an agent from scratch

    LabsA workflow; front-desk agent

  3. 11Sun Dec 20

    AI agents II: loop, tools, planning and approval

    The ReAct loop; designing tools; planning and approvals; agent frameworks

    LabsFramework agent; human approval for changes

  4. 12Sun Dec 20

    AI agents III: guardrails and reliabilityMini project demos

    Guardrails; retries and fallbacks; RAG for agents

    LabsHardened agent

Module 4MCP and the email agentJan 2–3
  1. 13Sat Jan 2

    MCP I: why MCP and your first server

    Why MCP; architecture; tools, resources and prompts

    LabsFirst MCP server; scheduling server

  2. 14Sat Jan 2

    MCP II: agents over MCP

    Clients and hosts; a knowledge server; control in the host

    LabsTwo servers, one agent

  3. 15Sun Jan 3

    MCP III: remote servers, security and deployment

    Remote MCP; MCP security; MCP in production

    LabsAttack and defend; deploy a remote server

  4. 16Sun Jan 3

    Project 2: the email agentProject 2 demos

    Discovery to goal; design; build; approval queues and outcomes

    LabsEmail agent build

Module 5Multi-agent systems, observability and evaluationJan 9–10
  1. 17Sat Jan 9

    Multi-agent systems and A2A

    When to use more agents; multi-agent patterns; the A2A protocol

    LabsSupervisor agent; billing agent over A2A

  2. 18Sat Jan 9

    The AI harness, agent memory and deep agents

    The AI harness; agent memory; deep agents

    LabsAgent memory; deep agent

  3. 19Sun Jan 10

    Observability: logs, metrics and tracing

    Logs, metrics and traces; OpenTelemetry; dashboards and alerts

    LabsTrace an agent; dashboard, alerts and bug hunt

  4. 20Sun Jan 10

    Evaluation I: judges, RAG and agent evals

    Evaluation fundamentals; datasets and graders; LLM-as-judge

    LabsAgent evaluations; judge calibration and RAG metrics

Module 6Production evaluation, security and costJan 16–17
  1. 21Sat Jan 16

    Evaluation II: production evals and reliability

    Production evaluation; reliability at scale; evaluation gates in CI

    LabsOnline evaluator; GitHub Actions quality gate

  2. 22Sun Jan 17

    Security, cost and performance

    Threat modelling (OWASP); red-teaming; model routing and caching

    LabsRed-team test suite; cost and latency benchmark

No class Dec 24–27 or Dec 31. The weekday cohort has class on Martin Luther King Jr. Day (Jan 18). Each class is two hours of teaching, followed by interview practice.

Before day 1

  • You are comfortable writing code
  • No AI experience needed: the course starts with how LLMs work
  • A computer with a camera and microphone: classes are cameras on
  • You live in the United States

Tools you’ll use

Agents
LangChain, LangGraph, LangSmith, Deep Agents
Tool and agent protocols
MCP (FastMCP, LangChain MCP adapters), A2A
Retrieval (RAG)
Chroma, Pinecone, pgvector, BM25 hybrid search, cross-encoder reranking
Observability
OpenTelemetry, Arize Phoenix, LangSmith tracing
Evaluation and release
LangSmith evaluations, LLM-as-judge, RAGAS, GitHub Actions
Models
OpenAI, Anthropic Claude, open models via Ollama and Hugging Face
Developer tools
Cursor, Claude Code, Streamlit, SQLite, Pydantic

02 · How every class runs

Two hours, the same four steps.

  1. 1

    Learn the idea

    Your instructor explains the concept and why it matters in real work. Ask questions as they come up.

  2. 2

    See it built

    Your instructor builds it step by step in front of the class and explains each choice.

  3. 3

    Build it yourself

    You do the session's lab in class while your instructor watches, answers and helps you fix problems.

  4. 4

    Interview practice

    Every class ends with interview practice. In the FDE course, that is 10 FDE interview questions each class.

Cameras on, never recorded, like a university classroom. Weekend classes run two sessions back to back; weekday classes run one session each evening.

Included with both courses:Live office hours + 1-on-1s24/7 help channelLCAE exam prep, fee back on a passCertificate of course completion

03 · FDE vs AI Engineer

What the two courses share, and what is different.

Only in the FDE course

  • Customer discovery and the full FDE arc
  • Advanced retrieval: hybrid search, reranking and RAG evaluation
  • MCP servers, remote deployment and MCP security
  • Multi-agent systems over A2A
  • Deep agents with memory and planning
  • Human approval queues for actions
  • Observability with OpenTelemetry, dashboards and alerts
  • Threat modelling and red-teaming

Shared foundation

  • How models work and choosing a model
  • Prompts as code and structured output
  • Tool calling and agent basics
  • Retrieval with citations
  • Evaluation basics
  • Each one rebuilt for that course’s project

Only in AI Engineer

  • Streaming chat and generative UI
  • Designing for AI mistakes and capturing feedback
  • Quotas, metering and personalization
  • Testing AI code in everyday CI
  • Feature flags and staged rollout
  • A/B testing prompts and models
  • Product analytics and data flywheels
  • Fine-tuning, distillation and running your own model
  • Voice, images and on-device AI

FDE graduates: about 16 of the 22 AI Engineer sessions are new material. Compare the two courses side by side.

Not sure which fits? Ask us live.

Free one-hour info sessions for each course, run by an instructor. Or answer 12 quick questions in the course finder.