Building intelligent applications that can interact with real-world data requires more than just Large Language Models (LLMs), it requires the ability to call external functions and tools. Tool calling transforms a conversational LLM into an agent that can execute code, query APIs, and perform tasks. In this blog post, we are going to create a job search assistant using the MLX-LM Server, connect it to the OpenAI client , and utilise the Qwen3-8B model’s tool‐calling abilities. We are going to build a tool that scrapes job postings from DEV.BG, a popular Bulgarian job board, and provides intelligent responses about available positions.

Most modern languages are written from left to right, thus we assume that thinking from left to right is the most natural way to process information expressed with these languages. This is particularly true for Large Language Models (LLMs) which are typically trained to predict the next word in a sequence, known as left-to-right (L2R) language models. But what if, for certain tasks, thinking backward could actually be better?

In this post, we explore the process of fine-tuning a language model for function-calling using MLX-LM. Following the Hugging Face Agents course notebook, we’ll walk through the steps from setting up the environment to training the model with LoRA adapters. The goal is to empower the model with the ability to intelligently plan and generate function calls, making it a versatile tool for interactive applications. Medium post can be found here

This blog post is going to be a tutorial on how to fine-tune a LLM with LoRA and the mlx-lm package. Medium post can be found here and Substack here.