The JavaScript AI Build-a-thon

4:00 CH - 5:00 CH (UTC)

Discover how to run powerful AI models entirely on-device using Foundry Local. We’ll cover the architecture, how to compile Hugging Face models for local execution, and how to integrate everything with LangChain.js. You’ll also see a compact Android demo showcasing low-latency, offline AI experiences. By the end, you’ll know how to set up, optimize, and build local-first AI apps, and how to join the private preview to go deeper. The Java Script AI Build-a-thon

4:00 CH - 5:00 CH (UTC)

In this session, we explore the end-to-end journey of designing and developing AI solutions with Microsoft Foundry models. The session is structured around four key phases of the model development journey. We begin with Model Understanding, where you discover the Foundry model catalog and build intuition around core model capabilities. Next is Model Selection, focusing on how to choose the right model for a given use case using benchmarks, evaluations, and leaderboards. We then move into Model Customization, covering how models can be adapted through techniques such as context engineering, fine‑tuning, and compression to better meet application needs. Finally, we close with Model Observability, highlighting how evaluations, tracing, and monitoring help ensure models behave safely, reliably, and effectively at scale. The Java Script AI Build-a-thon

5:00 CH - 6:00 CH (UTC)

This session breaks down advanced RAG patterns, from chunking and embeddings to similarity metrics, vector storage, retrieval strategies, and LLM chain responses. We’ll also explore streaming, chat history, and where GraphRAG fits in. You’ll leave with a clear blueprint for building accurate, optimized RAG pipelines for personal or enterprise-scale scenarios. Guided project: You’ll build a local serverless RAG chat app using CodeTours covering architecture, ingestion, Cosmos DB/FAISS storage, retrieval, and response generation. By the end, you’ll have a working repo plus practical intuition for tuning chunking, similarity, and evaluation. The Java Script AI Build-a-thon