Training a Bank Support Bot: Quality Data and Thousands of Intents
summary
Tinkoff’s support bot Oleg is trained with NLP models to handle over 1,000 distinct intents across products and customer-service topics. The talk covers building and labeling large-scale, high-quality training data and the engineering needed to train and deploy intent classification for a bank’s support workflows.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:32Automating support challenges
- 7:18Old paradigm problems
- 10:24Data platform
- 14:40Data labeling and quality
- 17:10Classification architectures
- 21:50Language model integration
- 24:41Summary
- 25:02Q&A
related talks
26mTinkoff Robot Battle: Voice Assistant vs Chat Bot
35mChat Bot Building Story - The Good, the Bad and the Unexpected
45mFinding the balance between human and bot interaction
13mWhy Chatbots Fail, And How To Fix Them
32mHarnessing Large Language Models for Enhanced Customer Experience
22mChatbot testing in the real world: Making sense of an emerging practice
37mIntelligent Customer Support for the Masses: How Nubank Does It
58mA Primer on AI Powered Conversational Experiences and Language Models