Conversational Agents for Job Recommendation

The goal of the project was to obtain an overall compatibility index between a resume and job offers, collected through scraping techniques from LinkedIn, in order to identify the positions best matching the candidate's profile.

The project involves the creation of conversational agents with the Autogen library that provide an overall compatibility value between the resume and the job offers, preliminarily selected through a similarity computation with the MPNet sentence transformer model. Each agent is responsible for a specific aspect of the resume, such as Hard Skills, Soft Skills and Salary; when salary information is missing, the dedicated agent produces an estimate.

The resumes come from a Kaggle dataset and required preprocessing operations, including Named Entity Recognition (NER) tasks for skill identification. Communication between agents happens through Mistral 7B quantized to 4 bits.