## Automated Generation of Q&A Datasets from Call Centers A new automated approach promises to simplify the adaptation of Large Language Models (LLMs) to specific domains. The proposed solution transforms audio recordings from call centers into high-quality datasets, ready for model fine-tuning. The pipeline, named Call2Instruct, automates the entire process, starting from audio processing (diarization, noise removal, automatic transcription) to semantic extraction of customer demands and attendant responses. Semantic matching via vector embeddings generates the final question-answer pairs. The generated dataset was successfully used for fine-tuning an LLM model based on Llama 2 7B, demonstrating the feasibility and practical value of the approach. The developed code is publicly available to promote reproducibility and future research. ## Implications for Customer Service The ability to convert unstructured conversational data into usable resources for LLM training opens up new perspectives for creating more effective artificial intelligence systems in the customer service sector. This could lead to faster, more personalized, and accurate responses for customers.