Biography
I am Head of Technology at Finster AI, where I lead the company’s AI and technology strategy and oversee the technical organization. I joined Finster as AI Lead on the founding team, building AI agents to fast-track financial decision making, before transitioning into this role.
Previously, I did a DPhil (PhD) in Machine Learning at the University of Oxford, advised by Professor Philip Torr in Torr Vision Group and Professor Yarin Gal in OATML, closely supervised by Dr. Puneet Dokania. I obtained my Bachelor of Engineering degree in Computer Science and Engineering from Jadavpur University, Kolkata, India and completed my Master of Science (MSc) in Computer Science from the University of Oxford.
My research background mainly focuses on methods to safely scale modern neural networks. I am passionate about building and deploying frontier AI safely in domains where reliability and trust are non-negotiable.
Interests
- Safe and scalable AI
- AI Agents
Education
-
DPhil (PhD) in Machine Learning, 2019-2024
University of Oxford
-
MSc in Computer Science, 2017-2018
University of Oxford
-
BE in Computer Science and Engineering, 2012-2016
Jadavpur University
Experience
Head of Technology
Finster AI
AI Lead, Founding Team
Finster AI
Research Intern
Meta AI
Research Intern
Meta AI
DPhil (PhD) Student
University of Oxford
Thesis: Methods to Safely Scale Modern Neural Networks. Research on:
- Uncertainty and Calibration of Deep Neural Networks
- Application in large scale computer vision problems
- Multi-modal foundation models
Research Intern
FiveAI
Graduate Student (MSc) in Computer Science
University of Oxford
Software Development Engineer (SDE)
Amazon Development Centre
Responsibilities included:
- Design and implement a method to support real-time data transfer from OLTP datastores to OLAP datastores.
- Design and implement APIs that can scale and support Amazon’s seller platform launching in new countries with minimal effort.
Software Development Intern (SDE-Intern)
Amazon Development Centre
Undergraduate (BE) Student
Jadavpur University
Featured Publications
Fine-tuning can cripple your foundation model; preserving features may be the solution
We analyze concept forgetting while fine-tuning foundation models and propose a simple fix to this phenomenon.
Raising the Bar on the Evaluation of Out-of-Distribution Detection
We propose a new benchmark for generating and evaluating different types of out-of-distribution samples given an in-distribution dataset.
Deep Deterministic Uncertainty: A New Simple Baseline
A deterministic deep neural network with sensitivity and smoothness (bi-Lipschitz) constraints on its feature space can be used to quantify epistemic uncertainty from an estimate of density in feature space and aleatoric uncertainty from the entropy of its softmax distribution.
Open Vocabulary Semantic Segmentation with Patch Aligned Contrastive Learning
We propose a modified contrastive loss function which allows training an alignment between patch tokens of a vision encoder and text CLS token of CLIP like models. This loss allows for easy seamless transfer to semantic segmentation without requiring additional annotations.