Work
Research
I'm a Senior Data Scientist at Mastercard's AI Garage, where I work on machine learning research for fraud and risk models trained on billions of transactions, along with the data problems that come with that scale.
My research sits at the intersection of representation learning and data-centric ML, across vision, graphs, and tabular data. What interests me is what happens when the data itself is imperfect, via noisy labels, severe class imbalance, scarce supervision or distributions that shift over time.
Some of this is theoretical curiosity: what a model memorizes versus generalizes, how neighborhood structure shapes representations, how learned embeddings behave when the underlying distribution is adversarial or long-tailed. But much of it is grounded in working directly with large-scale financial transaction data, where class imbalance is not a benchmark setting but a structural reality, where fraud patterns evolve adversarially, and where a model's failure modes have real consequences. That experience shapes how I think about research problems. I'm drawn to methods that are principled enough to publish and robust enough to actually deploy.
Publications
Towards Equitable Coreset Selection: Addressing Challenges Under Class Imbalance
Which data you train on matters as much as how you train, especially at the scale of billions of samples and even more so when that data is heavily imbalanced. This looks at when and why standard coreset selection fails under class imbalance, and proposes a more equitable approach that stays efficient.
AMEND: Adaptive Margin and Expanded Neighborhood for Efficient Generalized Category Discovery
Generalized category discovery asks a model to recognize known classes while also discovering novel ones in unlabelled data, which makes the neighbourhood you contrast against unreliable. AMEND expands that neighbourhood to recover more true positives, and adapts the margin per sample so the uncertain ones are treated more conservatively.
AdaPrompt: Prompt Tuning with Adaptive Neighbours for Generalized Category Discovery
Also on generalized category discovery, but through prompt tuning: adapting a pretrained backbone with a small set of learned prompts, guided by adaptive neighbours in the unlabelled set. Both papers are from my masters at IISc, advised by Prof. Soma Biswas. The open-world setting forces you to ask what a good representation even means when the label space itself is incomplete.
Study of Topology Bias in GNN-based Knowledge Graph Algorithms
Findings from work on knowledge graph representations with graph neural networks, during my internship at Mastercard.
Talks
The Trustworthiness of AI
On what trustworthy AI means in practice: reliability, interpretability, and the gap between benchmark performance and deployment.
Women in AI: Initiatives and Impact
Representing Mastercard AI Garage's work on building more inclusive AI research communities.
Career Journeys in AI (panel moderator)
A panel on career paths in AI, with a focus on non-linear journeys and underrepresented voices in the field.
Community
Mastercard AI Garage
I co-organize the Paper Reading Group, a weekly session on new and seminal work in AI/ML. Active member of Women@AIG.
Founding member of what is now a statewide initiative for female engineering students. Led it through its first year on campus, coordinating over a dozen sessions between students, alumni and industry experts.
WITI (Women in Technology International) ↗
A 30-year-old global network promoting women in technology. Program Manager for WITI India for six months, and co-organized their first APAC conference.