Sumanyu Ghoshal / sʊ-man-juː • ɡoʊ-ʃɑːl /

Building Voice AI Agents that help consumers in making correct, complex financial decisions. Fascinating to have AI Agents and Systems aligned in multi-turn conversations spanning complex financial (read- structured) information while enabling humans to truly understand their financial situation. Previously learnt engineering at Palantir and in Comp. Sci. and Engineering Courses during my undergrad at IIT Bombay

Email LinkedIn
Sumanyu Ghoshal

Experience

Product Manager (AI Workforce for Consumer Finance) , Prodigal
Nov 2023 - Present
  • Shipped ProAgent from scrappy prototype to enterprise rollouts; hardened our voice stack for PCI-grade workflows; built the automated evaluation suite that keeps every conversation compliant; unlocked multilingual deployments so regulated lenders can serve borrowers in their native language.
Voice AIAI AgentsPCI DSSGTM Strategy
Strategy Analyst , Prodigal
Nov 2021 - Oct 2023
  • Learnt how startups get built - from GTM to hiring incredibly talented folks.
Forward Deployed Engineer Intern , Palantir Technologies
London, UK | Summer 2021
  • Saw how the FDE model enables product development and customer delight by solving some of the most complex business and technical problems in one go. From PySpark pipelines to Java code - experienced working on the entire backend stack in summer. Also realized that Software Engineering is not something that I enjoy as much as pure product development roles - which pushed me into a PM role.
Palantir FoundryJavaPySpark

Education

Indian Institute of Technology Bombay
B.Tech w/ Honors, Computer Science
2018 - 2022 | GPA 9.26/10
✦ Institute Academic Prize (Top 2.5%ile)
✦ AIR 160 JEE Advanced

Coordinator, Dept Academic Mentorship Program. Managed 36 mentors. Redesigned the TA Application Portal.

Research

Learning2Select: Subset Selection , IIT Bombay & Google Research
2022
  • My attempt at a research project with one of my best friends. Worked with leading researchers in India around how a subset of data could be used to train models and reach similar accuracy as the entire dataset. By representing each ML model as a Computational Graph, and by creating a Function Approximator model that uses the Computational Graph as an input along with the dataset, we were able determine the most relevant sample keeping in mind the constraint of the % sample needed.

Beyond Work

Guitar, Trying to read books, Reading Fiction, Exploring new Cafes, Cities, Aeorplanes and Airports.