Highlights
- Build a stateful, tool-using agent loop (streaming, retries, timeouts, safety rails)
- Implement end-to-end RAG: ingestion, embeddings, retrieval, reranking, citations
- Create an MCP server with schemas, permissions, and real tools
- Add memory + workflows (ReAct, plan-and-execute, routing patterns)
- Upgrade to multi-agent systems (router, researcher, critic, executor)
- Production hardening: testing, evals, logging, retries, caching, deployment
- Ship a full capstone: chatbot + RAG + tools + multi-agent workflow
Course Details
Course kickoff + dev setup (code-first)
- What “agentic” means in practice: tools, memory, planning, orchestration
- Local environment: Python, venv/uv/poetry, lint/format, pytest, type hints
- Repo scaffolding:
apps/(agents),services/(MCP),libs/(shared)
Python fundamentals (only what we’ll immediately use)
- Data structures, modules/packages, CLI args, logging
- OOP vs functional style for agents (dataclasses, protocols/interfaces)
- Async essentials:
asyncio, tasks, queues, timeouts, cancellation - HTTP basics: FastAPI (or equivalent), requests/httpx, websockets (optional)
Agent core: a minimal “tool-using” loop
- Message format + state (conversation, scratchpad, structured outputs)
- Tool registry pattern (schema, validation, routing, error handling)
- Safety rails: allowlists, rate limits, redaction, audit logs
- Evaluation harness: golden tests, replay, deterministic mocks
MCP server (capabilities as tools)
- MCP concepts: tool discovery, schemas, calling conventions
- Build a minimal MCP server in Python (hello tool → real tools)
- Implement 3–5 practical tools:
-
- Filesystem (sandboxed), HTTP fetch (restricted), DB query (read-only), calculators, internal APIs
- Auth + permissions model (per-tool scopes), observability, versioning
Chatbot (the “front door”)
- Build a simple chat UI/CLI + API backend
- Streaming responses, tool call UX, citations/grounding display
- Session storage: in-memory → Redis/Postgres upgrade path
RAG fundamentals (grounded answers)
- Chunking, embeddings, vector stores (local first, then managed option)
- Retrieval strategies: hybrid search, metadata filters, reranking
- Prompting for grounded responses + citation linking
- RAG evaluation: recall/precision sanity checks, adversarial queries
Agent memory + workflows
- Short-term memory (session), long-term memory (user profile / notes)
- Summarisation, decay, and “don’t remember this” controls
- Workflow patterns: ReAct, plan-and-execute, toolformer-like routing
Exposing capabilities safely
- Capability design: input contracts, idempotency, side-effect isolation
- Human-in-the-loop approvals for risky tools
- Secrets management, policy checks, sandboxing, audit trails
Multi-agent systems
- When multi-agent helps vs hurts (latency/complexity tradeoffs)
- Roles: router, researcher, coder, critic, executor
- Coordination patterns: supervisor, swarm, debate, blackboard, task queue
- Shared context + shared tools; conflict resolution and consensus
Production concerns (as we harden the build)
- Tracing + structured logs, metrics, tool call timelines
- Reliability: retries, timeouts, circuit breakers, fallbacks
- Cost controls: caching, batching, model selection, token budgeting
- Deployment: containerise, run MCP + API, config per environment
Capstone project
- A full agentic app: chatbot + RAG + MCP tool suite + multi-agent workflow
- Demo scenarios, test suite, and “how to extend” checklist
Who should attend
This course is designed for technical professionals who want to move beyond prompt experimentation and build production-ready agentic AI systems.
Ideal for:
- Software developers comfortable with basic Python
- AI/ML engineers building LLM-powered applications
- Backend engineers integrating AI into products
- Technical architects designing AI-driven systems
- Startup founders and product engineers shipping AI features
- Data professionals expanding into agentic workflows
If you can write basic Python and want to build systems that use tools, memory, RAG, and multi-agent coordination—this course is for you.
Feedback
4.8 out of 5 average
"Our tailored course provided a well rounded introduction and also covered some intermediate level topics that we needed to know. Clive gave us some best practice ideas and tips to take away. Fast paced but the instructor never lost any of the delegates"
Brian Leek, Data Analyst, May 2022
