Start a conversation ↗

Description

MLOps and AI Agents: Industrialising Models, Orchestration and Monitoring

Models that perform well in the lab and fail to survive a single quarter in production

  • 1 day — 7 h
  • In-person or virtual
  • Expert
  • Up to 6 participants

Between a model that scores well on a test set and a model that makes decisions every day in production lies everything data teams discover too late: training runs that cannot be reproduced, input data that drifts, no rollback path, inference costs that escape control. The arrival of agents adds further uncertainty around actions triggered automatically.

A one-day programme designed for practitioners: the full industrialisation pipeline, model registry, deployment and monitoring, then the integration of agents into decision processes with the corresponding guardrails. The exercises are based on architectures brought in by the participants themselves.

Learning objectives

  • Design a reproducible industrialisation pipeline from data through to the deployed service
  • Set up a model registry and version management
  • Deploy a model and organise its rollback in the event of an incident
  • Monitor data drift and performance degradation in production
  • Integrate an agent into a decision process together with its guardrails
  • Quantify and control the running cost of an automated decision service

What makes this programme different

The architectures examined are those of the participants rather than demonstration diagrams
A drift detection workshop run on a deliberately corrupted dataset
A decision grid for choosing between deterministic processing and an agent depending on the target process

Programme

1The industrialisation pipeline

From analysis notebook to reproducible service

  • Version control for code, data and models
  • Automating training runs and tests
  • Model registry and promotion between environments
  • Packaging and exposing the inference service

2Running models in production

Keeping a model alive

  • Deployment strategies and rollback mechanisms
  • Technical monitoring and business monitoring of the service
  • Detecting data drift and performance degradation
  • Triggering retraining and deciding when it is justified

3Agents in decision-making

Automating without losing control

  • Scope entrusted to an agent and permitted actions
  • Orchestrating calls to models and tools
  • Traceability of decisions and logging of actions
  • Guardrails, human validation and emergency stop

Who is it for

Data scientists · machine learning engineers · architects and data team leaders responsible for putting models into production.

Prerequisites

Practical experience of modelling and development in Python as well as the fundamentals of containerised deployment.

Dates & locations

36 scheduled dates between November 2026 and December 2027. Seats are confirmed in the order enquiries are received.

November 2026

December 2026

January 2027

February 2027

March 2027

April 2027

May 2027

June 2027

September 2027

October 2027

November 2027

December 2027

None of these dates suit you? We open additional sessions on request, and any programme can be run privately for your team.

Practical details

Before the programme
Online positioning questionnaire. Your development objectives are shared with the trainer, who tailors the practical case studies to your context.
Teaching methods
Theoretical input, workshops and practical case studies. Digital course materials and method sheets provided.
Assessment
Multiple-choice tests and role-play exercises. Assessment of learning at the start and end of the programme, with immediate and 60-day follow-up evaluations.
After the programme
One year of access to the e-learning platform. Self-assessment of the skills acquired and a 30-day follow-up session with your trainer.
How to register
Registration online or on the basis of a quotation.
Lead time
11 working days after confirmation of registration.
Accessibility
Accessible to people of determination. Contact our accessibility coordinator to design a suitable solution: contact@mpf-academy.ae
Start dates
Rolling intake: in addition to the scheduled sessions, this programme can start on request.