Introduction

“How to Activate Your Knowledge” is a 12‑hour intensive module delivered over three mornings (09:00–13:00), designed for decision-makers, managers and business leads rather than developers. It explains how an organisation can turn its documents and internal expertise into reliable, traceable AI-assisted answers, and then into actions. Retrieval-Augmented Generation (RAG) remains the foundation of the training; on top of it, participants discover AI agents, tool use and the Model Context Protocol (MCP). The training combines clear concepts, live demonstrations and a guided application to a real business case. No coding is required.

Module Objectives 

Finality of the Module 

Enable decision-makers to identify, scope, evaluate and govern knowledge-activation projects, from a RAG assistant grounded in internal documents to agentic workflows connected to business systems, and to steer these projects with the right questions, budget and risk controls. 

Key Competencies Developed 

• Understanding how LLMs, embeddings and semantic search work, deep enough to challenge claims and ask the right questions without writing code. 
• Understanding the RAG architecture (ingestion, chunking, indexing, retrieval, generation) and the design choices that drive answer quality, cost and traceability. 
• Understanding what AI agents are, how they plan and use tools, and where they add value compared to a classic chatbot. 
• Understanding MCP (Model Context Protocol) as the emerging standard for connecting AI to internal data and applications, and its implications for architecture, security and vendor choice. 
• Ability to assess risks: hallucinations, data confidentiality, access rights, human oversight and regulatory requirements (EU AI Act, GDPR). 
• Ability to evaluate a solution or vendor proposal: quality indicators, total cost of ownership, build vs. buy vs. partner. 

Expected Outcomes for Organisations 

• A shared vocabulary between management, business teams and technical teams. 
• Ability to identify and prioritise high-value knowledge use cases within their own organisation. 
• Informed decisions on sourcing (build, buy, partner), hosting (local vs. cloud) and technology (open-source vs. proprietary). 
• A governance approach for deploying AI assistants and agents with traceable, reliable answers. 
• Reduced risk of costly pilots that never reach production. 

Module Structure 

• Day 1 – Foundations (1/2): From documents to answers. LLMs, embeddings, vector search and RAG, explained through live demonstrations. 
• Day 2 – Foundations (2/2) and Agentic AI: From answers to actions. AI agents, tool use, MCP, governance, risk and economics. 
• Day 3 – Business Case: Guided application to a real scenario, roadmap and decision note. 

Detailed Content 

1 Foundations: Retrieval-Augmented Generation 

• Why “activating knowledge” matters: the hidden cost of scattered documents and tacit expertise. 
• LLMs demystified: what they do well, their limits, and why they hallucinate. 
• Embeddings and semantic search, illustrated: how a machine “finds meaning” in documents. 
• The RAG pipeline step by step: ingestion, chunking, indexing, retrieval, reranking, generation, and which choices drive quality and cost. 
• Live demonstration: a low-cost RAG assistant answering with cited sources on real documents. 
• Measuring quality: relevance, faithfulness to sources, citation coverage, user feedback. 

2 From Answers to Actions: Agentic AI and MCP 

• From chatbot to agent: planning, memory, tool use and multi-step workflows. 
• Agentic RAG: agents that decide where to search, reformulate questions and cross-check sources. 
• MCP (Model Context Protocol): a standard “plug” connecting AI to document stores, databases and business applications, and what it changes for integration, security and vendor lock-in. 
• Architecture patterns: single assistant, specialised agents, human-in-the-loop validation. 
• Live demonstration: an agent that queries a knowledge base through MCP, drafts a deliverable and requests human validation. 

3 Governance, Risk and Economics 

• Data confidentiality, access rights, on-premise vs. cloud hosting. 
• Regulatory landscape: EU AI Act, GDPR, sector requirements; reference frameworks such as ISO/IEC 42001 and NIST AI RMF. 
• Human oversight and accountability: where a human must stay in the loop. 
• Cost drivers: licences, usage (tokens), infrastructure, maintenance; build vs. buy vs. partner. 
• Reading a vendor proposal: the key questions to ask. 

4 Confidential Use Case / Business Case 

Participants work in small groups on a real regulatory‑compliance scenario, with facilitators operating the tools. They will: 

• Frame a knowledge-activation use case: users, documents, typical questions, expected value. 
• Configure and test a RAG assistant that answers exclusively from official internal documents, with strict source citation and full traceability. 
• Extend the scenario with an agentic step (e.g. drafting, checking or routing) connected through MCP, and define where human validation is required. 
• Evaluate the results: answer quality, risks and limits. 
• Present a recommendation to a mock steering committee. 

Expected Deliverables 

Participants need to produce: 

• A use-case canvas: business problem, users, knowledge sources, expected value, success indicators. 
• A target architecture sketch (RAG, agents, MCP connections) at decision-maker level. 
• A risk and governance assessment: confidentiality, hallucination risk, human oversight, regulatory points. 
• A cost and sourcing estimate (build, buy or partner). 
• A short decision note for their management containing: 
• Context and business case. 
• Recommended approach and rationale. 
• Limitations and risks identified. 
• A phased roadmap: pilot, evaluation, scale-up. 

Participants can bring home: 

• A reference guide to RAG, agents and MCP concepts and vocabulary. 
• A use-case prioritisation grid to apply within their organisation. 

PRICE 

Thanks to the support of the European Commission and Innoviris in the framework of the EDIH sustAIn.brussels, SMEs and mid-caps receive this training free of charge (0€), in the context of de minimis aid. Large companies and participants without a company pay 960€ per participant. 

PRACTICAL INFORMATION 

• Language: English (bilingual exchanges FR/EN welcome) 
• Location: BeCentral, Cantersteen 12, 1000 Brussels 
• Format: In person, interactive, hands-on 
• Participants: Max 20 participants 
• Duration: 12 hours (3 days, 09:00–13:00) 
• Date: 16, 17, 18 Nov 2026 

Questions

Yavuz Sarikaya, Programme Manager at ULB

yavuz.sarikaya@ulb.be