Audience track
Production AI Agent Engineer
Build AI. Build agents. Build production systems. The most technical and most demanding of our tracks, for developers and technical engineers who want to go well beyond prompt engineering.
Positioning
Go beyond ChatGPT and basic prompt engineering
This track is for engineers who already write production code and want to build production-grade AI applications and agents: advanced RAG, stateful agent architectures, MCP and A2A interoperability, evaluation engineering, observability, guardrails and cloud infrastructure.
It is the most technical and most premium of our programs. The expectation is that you arrive fluent in Python and leave able to architect, deploy and operate an enterprise agent platform.
Who this is for
- Software developers
- Full-stack developers
- Backend developers
- Python developers
- Senior automation engineers
- SDETs with strong programming skills
- Technical architects
- Engineers transitioning into AI engineering
Eleven modules
Depth, module by module
| Module | What you learn | Stack |
|---|---|---|
| 1. Advanced Python | Advanced OOP, async programming, concurrency and type hints; Pydantic, API development, testing, packaging, logging and error handling. | Python · Pydantic · Pytest |
| 2. LLM engineering | Model selection, context engineering and structured output; function and tool calling, streaming, token and cost optimisation; model routing, open-source and local models. | OpenAI · Claude · Gemini · Ollama · Bedrock |
| 3. Prompt & context engineering | Dynamic context, memory, retrieval and tool context; long-context strategies, prompt versioning and context optimisation; the shift from prompt engineering to context engineering. | Context stores · versioning |
| 4. Advanced RAG engineering | Advanced chunking, hybrid search, metadata filtering and reranking; query rewriting, expansion and multi-query retrieval; parent-child retrieval, GraphRAG and agentic RAG; evaluation and production optimisation. | LangChain · LlamaIndex · pgvector · Qdrant · OpenSearch |
| 5. Production agent engineering | Agent architecture, tool calling, memory, planning and state; reflection, routing, supervisor patterns and multi-agent architectures; human-in-the-loop, long-running agents and failure handling. | LangGraph · Agents SDK · CrewAI · AutoGen |
| 6. MCP & A2A | MCP architecture, servers and clients, tools, resources and prompts; connecting agents to enterprise systems; A2A discovery, communication, delegation and interoperability. | MCP · A2A |
| 7. AI evaluation engineering | Offline and online evaluation; golden datasets and regression testing; LLM-as-a-judge, RAG and agent trajectory evaluation; tool-call, hallucination, performance and cost evaluation. | Ragas · DeepEval · LangSmith · Phoenix |
| 8. AI guardrails & security | Prompt injection and jailbreak defence; PII and sensitive-information filtering; output validation, tool permissions and agent authorisation; human approval, policy enforcement and OWASP LLM security. | Guardrails AI · NeMo · Presidio |
| 9. AI observability | Agent tracing, token tracking and latency monitoring; cost, prompt, model and evaluation monitoring; production logs, metrics and traces. | LangSmith · Phoenix · OpenTelemetry · Prometheus · Grafana |
| 10. Production infrastructure | Python, FastAPI, Pydantic, PostgreSQL and Redis; Docker and Kubernetes fundamentals; CI/CD with GitHub Actions or Azure DevOps; AWS Bedrock, ECS, Lambda, S3, RDS, IAM and CloudWatch. | AWS · Docker · Kubernetes · CI/CD |
| 11. AI coding & developer productivity | AI-assisted development, repository analysis, code generation and refactoring; debugging, unit and integration test generation, documentation and code review; agentic coding workflows. | Cursor · Codex · Copilot · Claude Code |
Capstone
Enterprise Autonomous Agent Platform
Build and deploy an enterprise-style agent platform containing an API gateway, supervisor agent, RAG agent, data agent, action and tool agent, MCP layer, evaluation layer, guardrail layer, human approval and full observability.
The final project must be deployed, tested, evaluated, secured and demonstrated as a production-style system rather than a simple chatbot. You present and defend the architecture the way a senior engineer would in a design review.
Deployed, tested, evaluated, secured and demonstrated.
Career outcomes
Senior and specialist AI roles
- AI Engineer
- GenAI Engineer
- AI Application Developer
- Production AI Engineer
- Agent Engineer
- AI Solutions Engineer
- LLM Engineer
- AI Technical Consultant
Questions
For developers and technical engineers
Real working proficiency. This track opens with advanced OOP, async programming, concurrency and type hints — it is not a Python course. If you are not comfortable writing and testing production Python today, start with the Launchpad or the AI-Powered Professional track instead.
Because stateful, long-running agent workflows with human approval and failure handling are what production actually demands, and LangGraph models that explicitly. You also get exposure to the OpenAI Agents SDK, LlamaIndex, CrewAI and AutoGen so you can compare and justify alternatives — which is exactly what an architecture interview asks you to do.
That is the requirement. The enterprise autonomous agent platform must be deployed, tested, evaluated, secured and demonstrated as a production-style system — not a notebook, and not a chatbot with a nice UI.
Depends what you are missing. If you have shipped LLM features but never built an evaluation harness, never traced agent cost in production, or never designed an MCP integration, there is substantial depth. If you have done all three, talk to admissions honestly and we will tell you if it is worth your money.
Next cohort
Build agents that survive production.
Talk to admissions about whether this track or the flagship better fits your engineering background.
Weeknight office hours · lifetime access to recordings · one accountable mentor across all 100 days