You don’t just finish with a certificate. You finish with a deployed AI system.

• 01EvalsCI-gated LLM evaluation. Golden datasets, LLM-as-judge, RAG-specific metrics, tracing. Wired into the pipeline so a quality regression blocks the merge. RAGAS, DeepEval, LangSmith, OpenTelemetry.
• 02AgentsMulti-agent orchestration, written from scratch. No framework. You build the tool registry, the observe-think-act loop, and every pattern by hand: routing, parallelisation, orchestrator-workers, evaluator-optimiser, reflection, handoffs, blackboard coordination, human-in-the-loop. Then map them onto the framework names for your CV.
• 03MCPModel Context Protocol, from primitives to a server you built. Tools, resources and prompts over stdio and HTTP, consumed by your own agent.
• 04ServingProduction inference. vLLM behind an OpenAI-compatible API, containerised, monitored, autoscaled, cost-tracked. API versus self-hosted decided with eval evidence, not vibes.

We could not find another live-mentorship programme anywhere that combines all four of these in a single curriculum. In Nigeria the picture is simpler: no programme publicly lists MCP, vLLM, or CI-gated evals at all.

• AI-assisted code review on every submission, 12 weeks
• Prompt-pattern and eval-harness template library
• Deployment boilerplate: Docker, FastAPI, vLLM, CI
• Reference implementations: worked RAG app, agent, MCP server
• Week 0 beginner bridge
• Alumni network and warm introductions
• Recorded session library, lifetime access
• Demo day slot plus recording
• Group CV and portfolio review

1. Before Week 1 · The Bridge

   For people starting from zero

   Two to three weeks, self-paced. Python, Git, the command line and pandas to the level Week 1 assumes, ending in an entry check. If you already code, skip it. If you do not, this is how you arrive able to keep up instead of drowning. Included free with your seat.

2. Weeks 1-4 · Phase 0

   Foundations

   Reproducible environments, real data cleaning, supervised learning framed properly. Classical ML with the maths just-in-time. Neural networks and training loops. Then transformers from the inside: attention, tokenization, embeddings, a forward pass traced from tokens to logits.

   Python 3.12 · uv · pandas · scikit-learn · XGBoost · PyTorch 2.x · Weights & Biases · Hugging Face

3. Weeks 5-8 · Phase 1

   LLM Application Engineering

   Week 5 prompting, context engineering and schema-validated output. Week 6 the whole of RAG in one week: chunking, embedding, hybrid search, metadata filtering, reranking, query rewriting, multi-hop and agentic retrieval. Week 7 wiring it into a business workflow with n8n, end to end. Week 8 evaluation and observability, with an eval suite wired into CI so a regression blocks the merge.

   Claude API · OpenAI API · Pydantic · instructor · LangChain · LlamaIndex · Qdrant / pgvector · n8n · RAGAS · DeepEval · LangSmith

4. Weeks 9-10 · Phase 2

   Agents and Protocols

   No frameworks. You write the orchestration yourself, in plain Python. Week 9 starts at the primitives, a tool registry, a run state, the observe-think-act loop, then implements every multi-agent pattern by hand: chaining, routing, parallelisation, orchestrator-workers, evaluator-optimiser, reflection, handoffs, blackboard, human-in-the-loop. You leave with your own agent library and a written comparison of what each pattern costs in tokens and latency. Week 10 adds MCP, a protocol rather than a framework, dropping straight onto the tool registry you built.

   Python · Claude and OpenAI tool use · MCP Python/TypeScript SDK · JSON-RPC 2.0 · no orchestration framework

5. Weeks 11-12 · Phase 3

   Serving, MLOps and Ship

   Serving an LLM with a high-throughput inference engine, containerised behind an OpenAI-compatible API, monitored and cost-tracked. Then integrating everything into one product and putting it in front of people.

   vLLM · FastAPI · Docker · Modal / RunPod · cloud GPU