LLM

Introduction The other day I was pondering on AI tooling and I thought about something that I didn’t know. How smart does an LLM actually need to be to reliably use function calling? At work, we take this for granted since we’ve been working on integrations with big models but that’s not always the case. Then, someone showed me an Instagram video about a workflow made with n8n, and I thought, does it have to be this way? It was a dead simple application, and it seemed a bit like a waste of a really powerful SaaS platform. Couldn’t this thing be made locally for free or at least for a fraction of the cost? What is the smallest model capable of performing these tasks without losing its mind? ...

Introduction Half a year ago, I started working in a different company. I was a bit stagnant on the previous one for different reasons, and ended up changing sector. Looking back, it was an intelligent choice, and quickly adapted to the new environment. I started doing AI-agent development and integrations, and I genuinely enjoyed the learning process. And what really piqued my interest was how AI was able to retrieve customized results there weren’t in the training data: Retrieval-augmented generation. ...