Microsoft Events presents an in-depth session on building and customizing local AI within enterprise apps using Windows AI APIs and Foundry, featuring speakers Connor Al-Joundi and Ivan Razumenic.

Integrating Local AI in Enterprise Apps with Windows AI APIs and Microsoft Foundry

Presented at Microsoft Ignite 2025 by Connor Al-Joundi and Ivan Razumenic

Overview

This advanced session walks developers through enabling local AI solutions on Copilot+ PCs using Windows AI APIs and Microsoft Foundry. The approach eliminates the need for cloud hosting and deep ML expertise.

Key Capabilities

  • Built-in Models (Phi Silica): Integrate native language and vision models for summarization, rewriting, and image understanding directly on Copilot+ hardware.
  • Turnkey AI Features: Activate features with minimal code, expediting AI-powered user experiences in enterprise applications.
  • Customization via LoRA: Use Low Rank Adapter (LoRA) techniques to fine-tune AI models with proprietary enterprise data, improving application relevance and accuracy.
  • App Content Search: Enhance applications with contextual search using AI to index and query app content.

Demonstration Highlights

  • Remote Manufacturing Challenge: Scenario-based demo illustrating real-world application and AI problem-solving.
  • RAG in Action (Root Cause Analysis): Demonstrates Retrieval Augmented Generation to improve actionable insights and output structuring.
  • LoRA Adapter Workflow: Step-by-step guide on loading and linking LoRA adapters during inference for custom intelligence.
  • Customer-Specific Instructions: Shows how to teach device-specific build instructions using RAG methods tailored to enterprise requirements.
  • Performance & Administrative Controls: Covers enhancements to the Phi Silica model and enterprise-grade controls for managing and auditing Windows AI APIs.

Getting Started

Chapters

  • Transition to demonstration scenario: remote manufacturing challenge
  • RAG in action for root cause analysis
  • LoRA for model customization
  • Loading LoRA adapter in inference
  • Teaching customer-specific instructions via RAG
  • Phi Silica performance improvements
  • Enterprise administrative controls for Windows AI APIs

Speaker Information

Connor Al-Joundi and Ivan Razumenic, Microsoft Events

Technical Concepts Covered

  • Windows AI APIs architecture
  • On-device inference workflows
  • LoRA fine-tuning for enterprise models
  • Integrating language and vision models (Phi Silica)
  • RAG for contextualized, actionable outputs
  • Security and administration best practices for enterprise deployment
  • Customizing enterprise apps with AI
  • Enabling AI-powered search and summarization
  • Improving performance and reliability of local AI workloads

Additional Resources