Large Language Models in Employee Recruitment Processes

This diploma thesis designs and builds an AI Matching Assistant, a proof of concept of a chatbot/agent system used directly by an employee considering internal mobility to obtain a shortlist of 3 to 5 most relevant open positions at a large Czech bank. The motivation is the scale of the internal-mobility agenda, which at the studied organization consumes a substantial share of the HR department's, management’s and individual employees’ capacity, on the order of tens of thousands of hours per year. The work delivers a language model and a working demo. Production implementation and integration into the bank's internal systems are out of scope. The thesis combines fine-tuning of an open-source encoder model JobBERT-V3 for the Czech-Slovak language space with a Retrieval-Augmented Generation architecture and a vector database. On an evaluation dataset, the resulting trilingual model outperforms the base model across all tested languages. Evaluation is limited to these offline retrieval-quality metrics.