CV

Post-Baccalaureate Fellow

Centre for Responsible AI (CeRAI), IIT Madras, India

• Supervisor - Prof. Balaraman Ravindran
• Leading the research and development of an explainable ensemble-based audio deepfake detection tool tailored to the Indian context, aimed at supporting fact-checkers, media organisations, and public-interest platforms.
• Chapter author of the Grounding Responsible AI in Agriculture, a project in collaboration with Meta OpenLoop. The work explores the current applications and associated risks of AI in agriculture, identifies key gaps in its development and deployment, and examines relevant policies, regulations, and legal frameworks. The paper emphasised the importance of responsible AI through actionable technical and policy recommendations, grounded in expert interviews and comprehensive research.

External Master's Thesis Student

Image Analysis and Biometrics (IAB) Lab, Department of CSE, IIT Jodhpur, India

• Supervisor - Prof. Mayank Vatsa & Prof. Richa Singh
• Created and curated a large (approx 1.3M samples) and diverse 26 generation techniques across 4 sets) bias-free benchmark multimodal multilingual deepfake dataset comprising identity-aware swaps and synthetic media.
• Performed qualitative experiments such as BRISQUE score for visual quality and FAD for audio quality. And benchmarked existing state-of-the-art unimodal (visual and audio) and multimodal deepfake detection models.

Bachelor of Science - Master of Science (BS-MS) Dual Degree

Indian Institute of Science Education and Research Bhopal

Major: Electrical Engineering and Computer Science; Minor: Data Science and Engineering

• Data Science and Artificial Intelligence - Machine Learning, Deep Learning, Natural Language Processing, Computer Vision
• Computer Science - Data Structures, Algorithms
• Mathematics - Linear Algebra, Probability and Statistics, Calculus

All India Senior School Certificate Examination (AISSCE)

Delhi Public School, Bokaro Steel City

Sciences Stream

• Physics, Chemistry, Mathematics, English, Information Practices, Physical Education

ILLUSION: Unveiling Truth with a Comprehensive Multi-Modal, Multi-Lingual Deepfake Dataset

The Thirteenth International Conference on Learning Representations (ICLR 2025)

Women Post-Baccalaureate Fellow

Centre for Responsible AI (CeRAI), IIT Madras

Qualified for Assistant Professor, Subject - CSA

University Grants Commission - National Eligibility Test (UGC-NET)

Interpretable Deep Learning for Text Classification

• Implemented an existing novel interpretability framework for text CNNs from scratch to linguistically motivated features. The proposed features are borrowed from existing work on text classification of shorter texts between fiction and non-fiction genres Brown Corpus).
• The exit criteria are to corroborate our interpretability results with the baseline results obtained from the existing work using logistic regression.

Consumer Complaints: Text Classification Problem

• Statistical and deep learning models were used to solve a multi-class nominal label classification problem
• A vectorization and feature selection pipeline was used for statistical architecture, and hyperparameters were fed to a cross-validation grid search on 5 supervised learning models. Logistic regression gave the best results, accuracy of 85.9%.
• For DLarchitecture, FFN, bi-directional LSTM, RNN, and transformers were used. Bi-directional LSTM gave the best results, with an accuracy of 86.5%.