AgentK: Machine Learning Chatbot - Aditi Agarwal

Published on: Monday, 16 September 2019

"“Know it all” chatbots are like “know it all” humans: talkative, but useless. In this talk, I will cover how the combination of three simple concepts creates a mind-blowing chatbot. We will go through the journey of building a simple, domain-specific and scope defined chatbot and see how much value it can bring to businesses.

The chatbot uses the principles of natural language processing and machine learning. The underlying tech stack is powered by Python, ELK, Scikit-learn and Pickle.

We will also discuss some real world chatbots and why they are good at doing what they do. Finally, I’ll provide some tips on getting started with building your own bot and things to look out for when your chatbot starts seeing user traffic.

Some of our key learnings when working with chatbots will include:
1. Speed of response from the chatbot as the data grows and the chatbot learns new concepts
2. Metrics to measure the usefulness of the chat responses
3. Mechanism to handoff to humans when appropriate"