New VDF AI Enterprise is now available on AWS Marketplace
Instructor-led course · Advanced

Production Agentic Systems: Multi-Agent, RAG and Governance

Production Agentic Systems is an instructor-led course for teams taking agents beyond a pilot. Over four live half-day sessions you design multi-agent networks, run private RAG at production quality, govern tools and skills, place human oversight and budgets, and prepare audit evidence, finishing with a capstone that earns the VDF AI Certified Agent Engineer certificate.

  • 4 live half-days
  • 6 modules + capstone
  • Remote or on-site
  • VDF AI Certified Agent Engineer
Apply for this course See the syllabus
Level
Advanced
Format
Live and instructor-led, remote or on-site
Length
Four live half-day sessions (3.5 hours each)
Audience
Solution architects, AI engineers and platform owners responsible for agents in production
Cost
Free for customers and partners; quoted for other teams
Certificate
VDF AI Certified Agent Engineer
Reply to applications
Within 2 business days
Labs
One after every module
Outcomes

What you will be able to do

  • Decompose a business process into a multi-agent network that can be operated and audited
  • Run private RAG at production quality, inside each user’s permissions
  • Govern tools, MCP servers and skills so capability does not creep
  • Place approvals, policies and budgets where risk and cost actually sit
  • Prepare the audit evidence a security or compliance review will ask for
Before the course

Prerequisites

Syllabus

6 modules and a capstone

The modules run across the four sessions. Each ends with a lab your team completes in a VDF AI environment.

  1. Designing multi-agent systems

    From a business process to a network of specialist agents that can be run, changed and audited.

    • Decomposing a process into specialist agents
    • Network design: agents, tools and decision steps
    • Versions and templates for repeatable networks
    • Turning a plain-language task into a network
    Lab
    Design and build a network for a multi-step process, then version it as a reusable template.
    Outcome
    A network design you can defend in an architecture review.

  2. Private RAG at production quality

    Retrieval that stays accurate, current and inside each user’s permissions.

    • Ingestion from document sources and databases
    • Chunking and vector index choices
    • Permission-aware retrieval
    • Testing retrieval quality before and after release
    Lab
    Build a retrieval pipeline over mixed documents and a database table, then test it against a question set.
    Outcome
    A retrieval pipeline with measured quality and enforced access.

  3. Tools, MCP and skills at scale

    Keeping what agents can reach under administrative control as the agent count grows.

    • The tool registry and per-role grants
    • Registering remote MCP servers
    • Creating, validating and versioning skills
    • Importing and exporting skills between environments
    Lab
    Register an MCP server, grant its tools to one role only and ship a validated skill that uses them.
    Outcome
    A tool and skill estate an administrator can inventory and revoke.

  4. Human oversight, policies and budgets

    Deciding where people approve, where policy blocks and where spending stops.

    • Approval steps in networks
    • Policies and budgets
    • Monitoring runs and handling failures
    • Designing escalation paths
    Lab
    Add an approval step and a budget to a network, then trigger both and read the run record.
    Outcome
    Oversight placed by risk, not added everywhere by default.

  5. Model routing and cost control

    Matching each step to a model that fits its difficulty, sensitivity and cost.

    • Smart model routing in networks
    • Local models alongside hosted ones
    • Choosing a model per agent
    • Reading usage to control cost
    Lab
    Route the steps of one network across a small local model and a larger one, then compare quality and usage.
    Outcome
    A routing approach that holds quality while cutting cost.

  6. Governance, audit and roll-out

    Everything a security or compliance reviewer will ask for before wide release.

    • Roles, workspaces and sharing
    • Audit evidence for agents, tools and networks
    • Separating development, testing and production
    • A staged roll-out plan
    Lab
    Assemble the audit evidence for your lab system and draft its roll-out plan.
    Outcome
    A system ready for review, with evidence rather than assurances.
Capstone

Capstone: a governed agentic system for your own process

Design and demonstrate a multi-agent system for a process from your organisation, combining a network, retrieval, governed tools and human oversight, and present its design, evidence and roll-out plan to a VDF AI engineer.

Certificate

VDF AI Certified Agent Engineer

Awarded to engineers who complete Production Agentic Systems and pass the capstone review.

  • Designing multi-agent networks for real processes
  • Operating private RAG with measured quality and enforced access
  • Governing tools, MCP servers and skills
  • Placing human oversight, budgets and audit evidence
Included

What your team gets

  • Four live half-day sessions (3.5 hours each)
  • A hands-on lab after every module
  • A materials pack: session slides and lab guides
  • A capstone review with a VDF AI engineer
  • The VDF AI Certified Agent Engineer certificate on passing the capstone
How to join

Apply, agree dates, learn

  1. Send the application below. It takes two minutes.
  2. We reply within 2 business days to confirm eligibility and agree dates.
  3. Your team receives the materials pack, then joins four live half-day sessions.

Also available: VDF AI API and Integration Engineering.

Apply for Production Agentic Systems: Multi-Agent, RAG and Governance

Free for VDF AI customers and partners. Other teams receive a quote after the application is reviewed. We reply within 2 business days.

Course *

Fields marked * are required.