Abstract: Generative AI has taken over the imagination of the world. In particular, its ability to generate code from natural language instructions is amazing. At the same time, if you want to build products that can reliably do it, it is far from ideal. In production use cases, its accuracy has been observed to be 25-30%. ThoughtSpot, a leader in the analytics industry, has been trying to solve this problem for the last decade for non-technical users so they can derive insights from data without learning coding or complex product workflows. With a combination of Machine Learning, Prompt Engineering, Distributed Systems, UX, and Product thinking, we can push the accuracy to the high eighties and make a usable product. In this talk, I will discuss some of the key design choices behind this product.
Bio: Amit is a Co-founder and CTO of ThoughtSpot, which is one of the leading companies in the Analytics software space. Prior to that, Amit led multiple Systems and ML engineering teams at Google that built some of the largest Machine learning infrastructures globally and built models critical to Google’s revenue growth. Prior to that, Amit was one of the founding engineers on the Bing team at Microsoft, where he implemented the infrastructure for web-scale graph algorithms as well as led a part of the ranking team for search from scratch. Amit received his Ph.D. in Computer Engineering from the University of Texas at Austin and a Bachelor of Technology in Electrical Engineering from the Indian Institute of Technology, Kanpur. Amit is also one of the authors of the popular algorithms book series “Elements of Programming Interviews”.