Blog & Articles
Insights, tutorials, and best practices from my experience in software development, AI, and cloud technologies
All Posts
The Ultimate Guide to Compiling Hugging Face Models for Foundry Local
I spent a weekend getting Llama-3.2-1B from Hugging Face running on Foundry Local and honestly, it's not just download and go. You need Microsoft's Olive toolkit to convert models to ONNX format, and there are gotchas the docs don't warn you about like the missing inference_model.json that silently breaks everything, or cache directory confusion that'll waste your afternoon. It's more friction than something like llama.cpp, but if you need cross-hardware support on Windows with NPU/GPU/CPU flexibility, this is the proper path to learn.
Building Local-First Web Apps: Integrating the Foundry Local JS/TS SDK
Microsoft's Foundry Local now has a proper JS/TS SDK, and the best part is it uses the same OpenAI SDK pattern you already know — just pointed at localhost instead of the cloud. I've been testing it and for offline apps, privacy-heavy clients, or just saving API costs during development, it's a real time-saver. It's still preview though, model selection is small, and there are rough edges like orphaned processes if your app crashes. My honest take: use it for your dev loop and prototyping, but hold off on shipping it to production until it matures.
Read More
No-Cloud AI: A Beginner’s Roadmap to Foundry Local Development
Microsoft quietly dropped Foundry Local, a tool that lets you run AI models right on your machine no Azure account, no API keys, no cloud bills. I've been using it for a few weeks and honestly, the setup is dead simple: one command install, pick a model, and you're chatting in your terminal. It's still in preview and the model catalog is limited, but for local development, air-gapped environments, or just prototyping without cloud headaches, it genuinely fills a gap nobody else was addressing properly.
What are Azure Machine Learning Pipelines? The Backbone of Modern MLOps
Azure Machine Learning Pipelines aren't just for running scripts; they’re about turning messy ML workflows into reusable, modular components that actually save you time and money. By caching steps that haven't changed, you stop burning through your Azure budget on redundant data prep, while the component-based setup lets your team work on separate parts without stepping on each other's toes. Just a heads up: make sure you're moving to the v2 SDK now, as the old version is hitting its sunset sooner than you think.
Training Predictive Models with Azure ML Designer
Azure ML Designer is a solid entry point for visual learners or teams with zero coding experience, turning complex machine learning pipelines into simple drag-and-drop flowcharts. While it’s fantastic for quick prototypes and explaining models to stakeholders, it quickly hits a wall when you need to handle serious MLOps, version control, or automated retraining. It’s a great teaching tool, but for production-scale projects, you'll likely save more time and money by skipping the canvas and sticking to the Python SDK.
The 2026 Guide to the Microsoft Foundry Tool Catalog
Microsoft’s latest rebrand to "Foundry Tools" might be confusing, but it consolidates heavy hitters like Speech, Document Intelligence, and the new Content Understanding into a single, modular ecosystem. The real value lies in features like Voice Live for low-latency agents and automated PII redaction, though the overlapping capabilities between services can make choosing the right tool a bit of a headache. Just watch your budget closely, those "cheap" individual API calls for translation or data extraction can snowball into a massive monthly bill once you hit production scale.
What is Foundry IQ? The Intelligence Layer of the Microsoft Ecosystem
Microsoft Foundry IQ is essentially a "RAG-in-a-box" service that handles the messy plumbing of enterprise data, like chunking, embedding, and syncing permissions, so you don't have to. Its standout feature is the deep integration with Entra ID and Purview, ensuring your AI agents actually respect document access levels without a custom-built security layer. While it’s a massive time-saver for anyone deep in the Azure ecosystem, keep in mind it’s still in preview, so expect some "new platform" friction before it’s truly production-ready.
The 2026 Guide to AI Agent Evaluation in Microsoft Foundry
Testing AI agents usually feels like "vibes-based" guesswork, but Microsoft Foundry’s new evaluation pipeline finally turns it into a real engineering discipline. By using automated evaluators for things like task adherence and coherence, you can catch tool-calling failures and instruction drifting before they hit production. Just keep an eye on the high token costs of "judge" models and the annoying regional limitations, it’s a powerful quality gate, provided you've got the budget and the right Azure region.
What is Microsoft Foundry Agent Service? The Future of Autonomous AI
Microsoft's new Foundry Agent Service is a solid attempt to simplify the messy world of AI agent infrastructure by handling the hosting, security, and scaling for you. While the platform shines with its enterprise-grade "on-behalf-of" authentication and deep integration into Microsoft 365, much of the best tech is still in preview. It’s a powerful shortcut for teams tired of building their own plumbing, but don't let the "no-code" promise fool you—managing costs and model quirks in production will still require some real elbow grease.