This session focuses on giving developers practical pathways to adapt AI models to their own data and domain needs. Discover the current tools, techniques, and best practices for making AI work in your context.

Agenda :

1.Welcome & objectives— set context, who’s in the room, and what you'll learn

2.From automation to agentic systems— evolution: automation → assistants → single agents → agentic systems

3.How we ask agents now— stream-of-thought, multi‑shot RAG, multi‑source questions; “one agent” architecture patterns

4.Retrieval stack deep‑dive— foundational RAG: keyword, vector, hybrid, reranking; semantic ranker and quality metrics

5.Foundry IQ: the knowledge layer for agents — knowledge bases, agentic retrieval, multi‑turn planning

6.Performance, monitoring & evaluations— metrics, automated evals

7.Model customization: RAG vs fine‑tuning — when to use each, hybrid strategies; RAFT and distillation patterns; stored completions

8.Fine‑Tuning Workflow in Microsoft Foundry - Types of Fine‑Tuning

9.RAFT and Distillation — Retrieval‑Augmented Fine‑Tuning - use a teacher model to generate synthetic Q/A data grounded in your document - Transfer knowledge from a larger teacher model to a smaller, cheaper model