Machine Learning-Enabled Chatbots
summary
Shows a working GCP pipeline that combines Dialogflow, App Engine, a Flask broker in Docker on Kubernetes, GPU TensorFlow instances and Mongo to serve search and QA over unstructured docs. Uses QANet on SQuAD-style data for machine comprehension, generates and fine-tunes domain data (about 20,000 question-answer pairs), and falls back to TFIDF when needed. Demo demonstrates intent/entity handling, partial live-demo failures with document retrieval, and practical tradeoffs for thresholds, testing and data labeling.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:34Chatbot Types
- 5:06The future with Robbie the robot
- 5:20Fathom bot
- 5:45The bottom line
- 11:43Challenges - Intent Detection
- 12:35Challenges - Entity Extraction
- 14:18Challenges - State Management
- 15:47Challenges - Information Retrieval
- 17:13Architecture Design
- 19:10Some tips
- 21:00Three information retrieval tasks
- 22:32Question Answering (QA) Models - Q20
- 24:16Reading comprehension is a harder task
- 25:06The future of reading comprehension Al2 Reasoning Challenge Sample Question
- 26:42Dataset augmentation & transfer learning
- 28:18Final considerations
related talks
46mChatbots, shall we shift into second gear?
44mGoogle Assistant powered by Containers, Machine Learning and NET on Google Cloud
58mA Primer on AI Powered Conversational Experiences and Language Models
51mA ChatGPT on your own data: how does it work?
57mBuilding smarter solutions with Machine Learning, from magic to reality
33mBuilding the next-generation Conversational AI with Python and Deep Learning
01h 12mBuild Your Own AI Chatbot with OpenAI & LangChain
47mBuilding smarter apps with Machine Learning, from magic to reality