Leveraging LLMs for Email Processing in Customer Centers

In this session, we present a project that we implemented in UKPN, which leverages large language models (LLMs) to classify per request, and urgency, summarize the contents of emails sent and generate an answer based on the available templates. The final solution is implemented with minimal changes to the regular business process and all of these results are fed back into the Outlook inbox. The solution utilizes Azure OpenAI, open-source models and MLFlow for model control and experimentation. We'll share the results, and decision processes, and compare the performance of different models and the costs of using these solutions.