Prashant Anand
Hi, I am working as a Machine Learning Engineer at Mercari, Inc. for the last 5 years in Tokyo, Japan. At Mercari, I work in the Customer Support domain and lead the exploration and application of ML, NLP, and LLMs in this domain. I work with Japanese text and so far I’ve developed features like sending automatic reply to customer inquiries using LLMs and MLMs, routing of customer inquiries, and template suggestion. Outside of work, I like exploring specialty coffee shops, photography, playing piano, reading (especially poetry), working out in the gym, and occasional hiking trips.
Session
Have you ever had a frustrating experience with a customer support chatbot? Issues like misunderstandings, irrelevant responses, and endless loops without reaching a human agent are common. So why do we use AI for sending automatic replies to customer inquiries at Mercari? Does it benefit customers? How do we ensure a positive experience?
Along with answering these questions, in this talk, we will look at the design of our highly precise system using machine learning for automatic replies. We will focus on Python content in sections about introduction to transformers, fine-tuning pre-trained transformer models on Japanese text, and the design and training of our ML model. Lastly, we'll discuss our A/B testing methodology and the impact on business metrics from using automatic replies.