Kaushik Chakrabarti

Leading development of Agentic AI solutions to revolutionize customer experience

Developing Responsible AI (RAI) guardrails for Microsoft's Generative AI products like Azure Open AI, Github Copilot and M365 Copilot. Shipped the following features in Azure RAI stack: • Models to detect hallucinations in Gen AI responses • Techniques to automatically measure quality of models that detect harmful content (like sexual, violent and hateful content) and generate precision/recall reports/dashboards for leadership • Solutions to detect abusive users and take action against them like throttling/suspending resources • Mechanisms to proactively detect AI threats for Microsoft's and competitor's products from external and internal sources. Started a widely circulated Microsoft-internal weekly newsletter reporting those threats.

Developed technologies to leverage knowledge graph to improve AI. Successfully shipped 2 features: • Leveraged business knowledge in customer databases like names of products and people in a compliant manner to improve speech recognition in business conversations in Dynamics Sales and Marketing. • Leveraged medical knowledge graph containing drug names and disease names to improve summarization of doctor-patient conversations in Nuance DAX Copilot. Both efforts led to significant improvement in the quality of speech recognition and summarization respectively.

Developed technologies for Natural Language-to-SQL and search over structured data. Shipped these technologies in Dataverse Search and Dynamics 365 for Marketing. Achieved top position in Natural Language-to-SQL leaderboard (WikiSQL) 3 times: for IncSQL in 2018, X-SQL in 2019 and HydraNet in 2020. People manager of NL-to-SQL team for 2 years.

Conducted research in database management, data mining, information retrieval, natural language processing and artificial intelligence. Published 100+ papers with total 11000+ citations. 17 papers with over 100 citations and 2 papers over 1000 citations. Inventor of 60+ Patents. Shipped research in numerous products like Bing, Cortana, Office 365 and Dynamics 365. Presented to executives including Bill Gates, Steve Ballmer and Satya Nadella on disruptive technologies. Served in program committees of top-tier database conferences like ACM SIGMOD, ICDE, VLDB, WWW, ICDM, ICME and journals like ACM TODS, VLDB Journal and TKDE. Served as associate editor of IEEE TKDE, member of the editorial board of Distributed and Parallel Databases Journal and PC vice-chair of ICDE 2012 conference. Mentored and managed 20 graduate students for summer internship projects, many of whom are now faculty members at reputed universities and scientists at top industry research labs.

Conducted research on approximate query processing and k-nearest neighbor search on complex high dimensional datasets. Won best paper awards at flagship database research conferences ACM SIGMOD 2001 and VLDB 2000.

Conducted research on providing approximate (instead of exact) answers to SQL queries to deal with huge data volumes and stringent response-time requirements of decision support workloads. Paper based on this work won the best paper award in VLDB conference and was invited to be published in VLDB journal.