Gestion des connaissances internes augmentée par l'IA

Ce qu'elle couvre

Ce programme enseigne aux équipes IT, RH et enablement comment déployer des pipelines de génération augmentée par récupération (RAG) sur des bases de connaissances d'entreprise comme SharePoint, Notion et Confluence. Les participants apprennent à concevoir des architectures de recherche sécurisées et respectueuses des droits d'accès existants, tout en améliorant considérablement la découvrabilité des connaissances. Le cours couvre la conception de taxonomies, les stratégies de découpage, la sélection de bases vectorielles et la conduite du changement pour favoriser l'adoption. Les sessions combinent ateliers d'architecture en direct et travaux pratiques avec de vrais connecteurs d'entreprise.

À l'issue, vous saurez

• Design a permission-aware RAG pipeline that ingests documents from at least one enterprise source and returns access-controlled answers
• Select and configure a vector store (e.g. Azure AI Search, Pinecone, Weaviate) appropriate to your organisation's data volume and compliance requirements
• Audit an existing knowledge base for RAG readiness and produce a remediation plan covering chunking strategy and metadata enrichment
• Define and measure at least three retrieval quality metrics (precision, faithfulness, answer relevance) using an evaluation framework
• Build and present a stakeholder adoption roadmap that addresses change resistance and tracks knowledge utilisation KPIs

Sujets abordés

• RAG architecture fundamentals: chunking, embedding, vector stores, and retrieval ranking
• Connecting enterprise sources: SharePoint Online, Confluence, Notion, and Google Drive via APIs and connectors
• Permission-aware retrieval: enforcing existing ACLs and role-based access in AI responses
• Data security and GDPR compliance for AI knowledge systems
• Taxonomy and metadata design to improve retrieval quality
• Evaluating and monitoring knowledge base freshness and answer accuracy
• Change management and employee adoption strategies for internal AI search
• Build-vs-buy decision framework: Azure OpenAI, Glean, Guru, and open-source stacks

Modalité

Delivered as a blended programme over 3–5 weeks: two live virtual workshop days (totalling 12–16 hours) plus async self-paced labs and readings (12–24 hours). An optional on-site implementation sprint day can be added for groups. Participants need access to a test instance of their enterprise knowledge platform. Labs are provided as pre-configured Jupyter notebooks and low-code workflow templates. Live sessions run in cohorts of 6–20; larger organisations can license the self-paced track for unlimited internal seats.

Ce qui fait que ça marche

• Engage a cross-functional steering group (IT, Legal, HR, a business champion) before any technical build to align on governance and scope
• Run a pilot on a single high-value knowledge domain (e.g. onboarding docs or IT helpdesk) before scaling to the full intranet
• Instrument the system from day one with retrieval quality metrics and user feedback loops to continuously improve answer relevance
• Pair the technical rollout with explicit communication about what the AI can and cannot do to set accurate employee expectations

Erreurs fréquentes

• Ingesting all documents without permission filtering, creating data-leakage risks where employees retrieve content above their clearance level
• Treating RAG as a one-time setup rather than an ongoing pipeline with freshness monitoring and re-indexing cadences
• Neglecting metadata and taxonomy work upfront, leading to poor retrieval quality that erodes trust within weeks of launch
• Launching without a change management plan, so employees revert to manual search and the system goes unused

Fournisseurs à considérer

• DataScientestdatascientest.com →
• Jedha Bootcampwww.jedha.co →
• Turing Collegewww.turingcollege.com →
• DeepLearning.AIwww.deeplearning.ai →

Sources

• Microsoft Azure — Implement RAG with Azure AI Search →
• Atlassian — AI in Confluence knowledge management →