Detailed Curriculum
Thirty-two weeks, four phases, two tracks running side by side.
Every week you build something technical and practise the consulting judgement that gets it bought, scoped and defended. Each phase returns to the same delivery questions with more demanding constraints.
Foundations
Goal: learn the language and working habits of production AI.
By week 4 you can cost a model, call one through an API with structured output, build a deterministic chain, and run a discovery interview that ends in a RAPID sheet.
LLM foundations-1. Build the working mental model: tokens, context window, temperature, hallucination.
Vibe coding-1 and -2. Build a backend on Supabase and ship the frontend on Vercel.
Hackathon-1. Build in your pod over the weekend, judged on a working deploy.
Map your skills against the FDE competency matrix.
AI landscape-1 and -2. Place any client request on the GenAI map; cost a feature from tokens and calls per day.
Prompt engineering-1. Structure prompts with roles, examples and JSON output; first calls through the OpenAI API and OpenRouter.
Read a sponsor, a champion, a user and a blocker, and write the update each one needs.
LangChain-1. See what makes a system agentic and run a first chain.
LangChain-2. Compose with LCEL and Runnables; enforce structured output with fallbacks.
LangChain-3. Build a multi-step chain on a business case, with retries.
Run a SPIN discovery in pairs, recorded and debriefed.
Workflow patterns-1 to -3. Route, parallelise and self-correct; pick patterns with the selection matrix.
Turn a vague brief into requirements, assumptions, priorities, implications and dependencies.
A first working AI app, a costed model choice, a multi-step chain and a RAPID sheet from a live discovery interview.
Core Build
Goal: move from individual tools to a dependable, deployed application.
By week 12 you can ship a traced, guardrailed agent on AWS AgentCore, ground it with retrieval, and write the PRD, estimate and priced proposal that go with it.
Mini Project 1. Build a multi-pattern workflow on an ops case and defend each pattern choice.
Score an organisation's AI readiness on five dimensions.
Tool calling-1 and -2. Turn client APIs into tools the model can call safely.
Memory-1. Add conversation memory with trimming and summarisation.
Lead a 45-minute simulated engagement with a difficult client.
Observability-1. Trace every run in LangSmith with cost and latency.
Guardrails-1. Validate inputs and outputs; learn the guardrail-as-a-tool pattern.
Evals-1. Build a golden set and prove one change with numbers.
Write a PRD with user stories and acceptance criteria.
AgentCore-1 and -2. Learn the AgentCore primitives and deploy the agent as a service.
The loop, assembled. Review the six parts of the agentic loop as one system.
Harden acceptance criteria and defend the PRD in a stakeholder review.
Mini Project 2. Ship the full loop on AgentCore: tools, memory, tracing, guardrails.
Self-paced. Cover deep learning basics and embeddings before the RAG block.
Sketch a solution on the Solution Architecture Canvas before any code.
Mini Project 2. Polish, deploy and record; join the open review of common failure patterns.
Defend an architecture in a trade-off tribunal.
RAG-1 and -2. Build a RAG chain with loaders, a vector store and citations.
RAG-3. Use Bedrock Knowledge Bases and S3 Vectors for managed retrieval.
Estimate five AI projects and see where the estimates diverge.
Enterprise RAG-1 to -3. Design ingestion for real corpora; compare chunking strategies and recommend one.
Price one project under fixed, time-and-material and outcome models.
Two shipped mini projects, a traced and guardrailed agent on AgentCore, a PRD, an estimate and a priced proposal.
Deployment Depth
Goal: work through the conditions in which promising AI systems usually fail.
By week 20 you can measure an enterprise RAG pipeline, run an agent on LangGraph with durable memory, and hold a scope through a hard client conversation.
Enterprise RAG-4 to -6. Design the pipeline, then build ingestion, indexing, retrieval and generation in two guided sessions.
Build an ROI model a CFO would sign.
Mini Project 3. Plug the pipeline into your agent so it answers with citations, deployed.
Map a Client Zero unit into bounded contexts within a week.
Advanced retrieval-1 to -3. Add knowledge graphs, hybrid search and reranking; measure each against the baseline.
Answer the seven objections to AI projects, recorded.
RAG evaluation-1 and -2. Score faithfulness and recall with RAGAS; fix the worst offender.
Enterprise RAG-7. Enrich chunks with metadata and filter at query time.
Draft Proposal 1 and take it through peer review.
LangGraph-1 to -3. Model agents as graphs with typed state; rebuild the week 4 patterns.
Plan pilot, scale and optimise with a change plan.
Agent memory-1 to -3. Checkpoint, resume and replay; keep memory across threads with a store.
Write an SOW with AI-specific clauses.
LangGraph-4. Compose subgraphs without tangling shared state.
Agent memory-4 and -5. Build a support agent that remembers a user across threads under retention rules.
Hold your SOW through a live scope creep simulation.
Mini Project 4. Rebuild your agent on LangGraph with durable state and long-term memory.
Score project health; review your portfolio and resume.
A measured enterprise RAG pipeline, an agent on LangGraph with durable memory, Proposal 1, an SOW and a delivery confidence score.
Capstone Engagement
Goal: run one engagement end to end and defend it in front of a panel.
By demo week you can run one engagement end to end: MCP tools, policies, a production deployment, a multi-agent design and an SLM you trained and served.
Mini Project 4. Deploy, record and defend the migration in a design note.
Design the team, then redesign it with 40 percent less budget.
MCP-1 to -3. Build an MCP server, test it with the inspector, and wire it into your agent with enterprise scoping.
Build a WBS, a RACI and a sprint plan for an AI agent project.
Observability-2 and Evals-2. Trace graphs and run regression evals on every change.
Guardrails-2. Add validation nodes, moderation checks and human-in-the-loop interrupts.
Resolve the three-way standoff between CTO, VP Ops and CFO.
Guardrails-3. Defend against prompt injection using the OWASP LLM Top 10.
Policies-1 and -2. Enforce role, spend and approval policies as middleware; run failure drills.
Deliver bad news and close a project out, recorded.
Platform governance-1. Layer Bedrock Guardrails over framework guardrails.
Deployment-1 and -2. Take the full pipeline live on AWS with a runbook.
Distil a programme into one board slide that earns a decision.
Project 5. Deploy a LangGraph agent with MCP tools, guardrails, policies and observability against a production checklist.
Own a full engagement in one simulation; write case study 1.
Multi-agent-1. Choose a topology, diagram first.
CrewAI-1 and -2. Build a crew with tools and delegation; sketch its LangGraph supervisor twin.
Draft five case studies and your portfolio narrative.
Multi-agent-2. Use councils and LLM-as-judge with their biases known.
Agno-1 and -2. Rebuild the case in Agno; score LangGraph, CrewAI and Agno.
Run the mock interview gauntlet and negotiate with a BATNA.
SLM-1 to -3. Match deployment topology to data residency; select a small model on task fit.
Defend the AI transformation workbook and brief the capstone client.
SLM-4 to -6. Fine-tune with LoRA and QLoRA; pair the model with retrieval in the RAFT style; serve it.
Deliver an update under pressure when the requirement changes.
Project 6. Train, tune and serve your own SLM; compare it with a hosted baseline.
Rehearse the panel and finalise the case studies.
Loop and harness engineering. Learn the two closing concepts and apply them to your scope.
Sprints 1 to 3. Freeze scope, build the core, harden and rehearse, with random-pick reviews each week.
Work with the consulting mentors in the sprint reviews.
Demo dry-runs, Demo days 1 and 2. Present a live system to the evaluation panel.
Complete final mock interviews and the consulting readiness evaluation.
Projects 5 and 6, five written case studies, a portfolio narrative, and a capstone system demonstrated live to the evaluation panel.
Ready to apply?
Leave your number and an admissions lead will call you back to walk through eligibility, cohort dates and fees.