Insights
AI implementation insights, software development best practices, and technology trends
In this blog, James shares his hackathon experience of building an image classifier under tight deadlines, starting with ML.NET's transfer learning but later pivoting to ONNX for better accuracy and speed. He explains how ONNX models seamlessly integrate into .NET apps, showcasing ML.NET's real strength in interoperability.
LLMs are weird, even between different model versions. A story about how changing from GPT-4o to GPT-4.1 caused a subtle but critical bug due to literal interpretation of parameter descriptions.
We're in a new era of product development where 'vibe coding' is having a moment. AI agents help founders spin up MVPs quickly, but once you find traction, everything changes. Here's when you'll want a real dev team.
There's a kind of elegance in building AI that knows when not to answer. In theory, large language models can do anything. In practice, what makes them truly useful, in the hands of real people, in real systems, is not just what they can say, but how they respond when the path forward isn't clear.
Let's be honest, writing dev tickets isn't exactly the most exciting part of building software. But it's essential. A clear, well-detailed ticket can be the difference between a smooth release and a dev pinging you with, 'What's this supposed to do again?' That's why I started using ChatGPT to help me write them.
Reflects on why legacy technologies endure and how old-school skills remain surprisingly valuable.