From a language model in your app to a custom model running locally on-device. I choose the technique that works best per task, not the one that's most in the news.
Machine learning is a part of AI. When people say "AI" now they usually mean generative AI, like language models. For the choice in your product, that difference matters.
Small, fast and cheap: classification, prediction or recognition based on your own data. Ideal for on-device speech or image recognition.
Large language models that understand and write text: summarising, answering questions on your own knowledge, turning free text into structured data.
Often the best solution is a mix: a language model where flexibility is needed, a small ML model where speed, cost or privacy matter.
Task first, then technique. That way you avoid a heavy language model where a small model would do, or the other way round.
What exactly should the feature do, for whom, and when is it good enough?
Machine learning, a language model, or a combination, tested on examples from your practice.
Via your API, in your app, or on-device, with a fallback if the model isn't sure.
Quality, speed and cost stay visible after launch.
Tell me what the feature needs to do. I'll show you which technique fits.