Architecting for ML.NET & Cognitive Services
summary
Shows concrete architecture rules: keep ML outside core business logic with an interface like IForecastCalculator so models run as ports and adapters and remain unit-testable; demo uses Microsoft Cognitive Services for speech and ONNX models (examples 34 MB and 250 MB) run locally with the ONNX runtime or via MXNet/Azure ML; notes softmax output mismatches in C# during local ONNX inference; advises build vs buy vs integrate decision, and choose local vs remote deployment based on latency, hardware acceleration and stability.
AI-written summary · about
chapters (10)
- 0:00Introduction and welcome
- 4:08Architecting for AI overview
- 8:03AI in enterprise reality
- 12:17Enhancing user experience with AI
- 22:22Driving business decision-making
- 23:46Ports and adapters for AI models
- 30:48Build, buy, or integrate AI models
- 34:24ONNX and model deployment
- 42:18Deployment considerations
- 44:45Key takeaways and Q&A
related talks
50mA Pragmatic Deep Dive into ML.NET
08mInterview Speaker
41mAI + Machine Learning for .NET Developers
46mML.NET for developers without any AI experience
01h 04mIntroduction to ML.NET
53mBuilding Intelligent .NET MAUI Apps with ML.NET
52mML.NET - Machine Learning made for .NET
27mN Ways to Do Machine Learning on .NET