An open-source toolkit for vision and video mechanistic interpretability, with 75+ models, pre-trained sparse autoencoders, activation caching, and analysis tools.
I am a Research Engineer on the AI Search team at ServiceNow in Toronto. My work focuses on improving the search relevancy by evaluating differnt retrivel, reranking methods.
Previously, I worked in Multimodal and Document AI team at ServiceNow, for evaluating and improving the document extraction flows. I completed an M.Sc. in Computer Science (Machine Learning) at the Université de Montréal, where I was affiliated with Mila. My interests include machine learning, NLP, information retrieval, computer vision, and multimodal learning.
September 2025Joined ServiceNow in Toronto as a Research Engineer on the AI Search team.
June 2025Prisma was presented orally at the Mechanistic Interpretability for Vision Workshop at CVPR 2025.
May 2024Our paper, Transformer-based Joint Modelling for Automatic Essay Scoring and Off-Topic Detection, appeared at LREC-COLING 2024.
August 2023Our paper, Unsupervised Out-of-Distribution Dialect Detection with Mahalanobis Distance, appeared at INTERSPEECH 2023.
April 2023Published the SHL Labs blog post The Battle of the Pens: Insights into Detecting AI-Generated Text.



My work spans machine learning, NLP, information retrieval, document intelligence, and multimodal learning.
An open-source toolkit for vision and video mechanistic interpretability, with 75+ models, pre-trained sparse autoencoders, activation caching, and analysis tools.
An unsupervised transformer method that jointly scores essays and flags off-topic responses with topic regularization and Mahalanobis scoring.
An unsupervised OOD detector that uses intermediate wav2vec 2.0 embeddings and Mahalanobis distance to flag unseen dialect inputs.


UTG Llama Award, ServiceNow (Q3 2025).
Kaggle Silver Medal, RSNA-MICCAI Brain Tumor Radiogenomic Classification.
Mind to Market Innovation Challenge, award for a drone-based surveillance proposal.