At THINK 2026 we showcased the Granite 4.1 model family. The goal: show the whole is greater than the sum of its parts.
Speech-to-text, language models, and adapters, composed through Mellea into one real-time voice app, validated live.
Here's how: mellea.ai/blogs/every-layer-…
Using #mellea skills compiler to better manage the deluge of github notifications: psschwei.com/post/blog-melle…
Had a blast presenting Mellea at NY Tech Week!
We explored how to build reliable AI applications and how to make small models rock: ibm.biz/small-models-rock
One key theme: smaller, well-tuned models can deliver real-world reliability without massive scale (and cost).
A blog on how to get frontier-level results from small language models by using Mellea and Granite Libraries do the heavy lifting: mellea.ai/blogs/small-models…


