Agents think , let's make them act : Integrating open-weight LLM into a system for autonomous deterministic operation

Let's break down how your favorite AI works and then build our own AI system that works for us.

I had a painful task , converting 8000 job vacancies into a dataset. I could not afford to use online APIs and manual entry promised to be tedious so I pulled the developer move. Writing a script for that. The talk sheds light on how to integrate LLMs into existing systems and making them work for you. This talk will start with a simple function executing a web scrapping task and feeding a web page into the LLM to single out the needed data. Then using Qwen 2.5 and ollama we will then move to get user input to call a tool which gives us a data entry for our dataset.

Whilst the task will be an anchor point , the key take away will be the architecture and approach behind and the the few key things to consider when doing the integrations. To top that at each stage , I will highlight the importance of creating closed-system AI integrations , share some approaches I used and pitfalls encountered in the process. At the end of the session , I hope to leave the audience with an understanding of how easy it is to use Python for them to create their 'own' AI , with unlimited tokens and how to ensure there is fewer hallucinations.