Train

04 — Senior engineers & AI teams

Agentic AI training — LangGraph, CrewAI, RAG and MCP

The deep specialisation: graph-based agent architecture, multi-agent crews, agentic retrieval, MCP integration and the LLMOps to keep it running.

For
Senior engineers & AI teams
Modules
6
Delivery
On-site or remote
Length
Scoped per cohort
Request a briefHow these run

Why this programme

A single agent handles a single job. The moment a process has branches, retries, approvals or state, teams discover their agent was really a very expensive if-statement.

This programme is about the architecture that survives that discovery — the same patterns I use building multi-agent systems at Signify.

Curriculum

What it covers

  1. Agentic workflows with LangGraph

    Graph-based control flow, state, checkpoints and human-in-the-loop. What makes a multi-step agent reliable rather than impressive once.

    • LangGraph
    • State
    • Human-in-the-loop

  2. Multi-agent systems

    Crews, delegation and specialisation in CrewAI. When more than one agent genuinely helps, and when it just multiplies your failure modes.

    • CrewAI
    • Delegation

  3. Agentic RAG

    Retrieval that reasons about what to fetch next, evaluates what it found, and knows when to stop.

    • RAG
    • Retrieval
    • Evaluation

  4. MCP and system integration

    Wiring agents into real systems through the Model Context Protocol — the work that turns a chatbot into something operationally useful.

    • MCP
    • Integrations

  5. LLMOps in production

    Observability, evaluation harnesses, regression detection, latency and spend. Knowing an agent has degraded before a customer says so.

    • LLMOps
    • Monitoring
    • Evaluation

  6. Judgement: where agents do not belong

    The module that saves the most money. Which problems deserve an agent, and which deserve a function.

    • Architecture
    • Cost
    • Risk

What the team can do afterwards

  • Design an agent architecture with defensible boundaries
  • Ship a multi-step agent against your own systems
  • Instrument agents so regressions surface early
  • Control cost and latency in agentic workloads
  • Argue confidently for the simpler solution when it is right

Who it’s for

  • Teams building agent-based features
  • Engineers who have outgrown single-prompt solutions
  • Platform teams standardising agent architecture
  • AI teams moving prototypes into production

Prerequisites

  • Working knowledge of Python
  • Some prior LLM API experience helps but is not required

Delivered personally by Viren Gajjar

Questions

About Agentic AI Engineering

  • How is this different from the developer programme?

    Advanced AI for IT Professionals is broad — integration, performance, security across the whole surface. This one goes deep on agent architecture specifically, and assumes you have already shipped something simpler.

  • Do you cover evaluation properly?

    Yes, and it is usually the most valuable module. Most teams can build an agent; far fewer can tell whether a change made it better or worse.

Ready to scope a cohort?

Tell me the team size, their starting point and what you need them to be able to do afterwards. You’ll get a proposed shape and a fixed quote back.

Request a briefAbout Viren