.NET Rocks! Local AI Models with Joe Finney
11 snips
Oct 2, 2025 Joe Finney, a mobile product owner and Microsoft MVP, dives into the world of local AI models. He explores the capabilities of models from platforms like Hugging Face for tasks like OCR and image recognition. Joe explains the importance of Windows AI APIs for developers, the nuances of managing ONNX models, and the trade-offs between local and cloud solutions. He emphasizes privacy and cost benefits of local models while discussing hardware requirements and practical tools to experiment. A must-listen for anyone interested in machine learning on their own devices!
AI Snips
Chapters
Transcript
Episode notes
TextGrab Powers Local OCR
- Joe built TextGrab, which powers PowerToys' text extractor and performs on-device OCR.
- He moved from Windows OCR APIs to Tesseract for better accuracy despite added setup complexity.
Start With Windows AI APIs
- Use the new Windows AI APIs as the easiest entry point for local AI in Windows apps.
- Check device support at runtime so you avoid bundling large models or managing downloads.
Many Useful Models Beyond LLMs
- OCR, image segmentation, object detection and many task-specific models exist outside LLMs.
- Hugging Face hosts a vast variety of ready models for non-language tasks.