I'm currently a Software Engineer at Google.
In the Applied AI team, we are building Conversational AI products leveraging state of the art Large Language Models for Agentic Commerce and Customer Experience.
Selected Projects
-
Agentic Commerce with self improvement
Building commerce agents for the masses. Focussing on production pipelines for self-building, self-evaluation and self-improvement. Bringing alignment between real and synthetic datasets. Guided by Sercan Ö. Arık and Venu Vemula.
-
Conversation Summarization with long context
Brought the power of long context window from LongT5 models to conversation summarization at scale. Supervised by Yong Yao.
-
Knowledge Search with conversation context
Built powerful search backend using new hybrid strategies that work with explicit queries as well as with context from the conversation. In collaboration with Siamak Shakeri, Yinfei Yang and Ning Jin.
I mainly worked with AI Apps and AI for CX teams.
Selected Projects
-
Machine Learning platform and workflows
Led the development of end-to-end Machine Learning workflows for sales and their deployment on Kubernetes.
Contributed to the Machine Learning platform, simplifying how models are trained and served.
-
Sales Product Recommendation quality
Brought over 11% accuracy improvement using novel techniques. Presented in Oracle Machine Learning Summit.
-
Knowledge graph from source code
Created knowledge graph of entities extracted from source code to enhance discovery. In collaboration with Ron Viscuso.
At eBay Inc I contributed to Apache Kylin, an open source Distributed Analytics Engine from eBay Inc, re-routing the queries not supported by Apache Kylin to Hive on Spark.
This gave me exposure to some of the most important technologies & services in distributed systems like Hadoop, Spark, HBase, Hive, Kylin, Ambari, HDP, Zookeeper, Tomcat, etc.
This is also where I met one of my best friends, Docker, who was a 6 month old kid back then. :)
At connecto.io, I developed a scalable web crawler using MEAN stack to find add-ons used by millions of websites.
We also predicted engagement scores for websites by analyzing features like visibility of call to action, contrast ratios and link densities using OpenCV library in Python.
This gave me exposure to the power of MEAN stack and also to an interesting domain of image processing.