Researcher
My current research explores Graded Algorithmic Inequality — how caste-based discrimination is quietly reproduced across Large Language Models (LLMs), text-to-image systems, and broader AI pipelines. By systematically testing these models, I investigate how they mirror historical social divisions: privileged identities are often associated with professional spaces and authority, while lower-caste profiles are linked to menial labour and impoverished environments.
I am especially interested in how bias hides and compounds through proxies — surnames, dialect, and institutional records — as it travels across an AI system, from early pre-training data through to automated decisions and multi-step agent workflows. For detailed experimental results and open-source materials, please see my GitHub repository.
To ensure AI does not become a new instrument of exclusion, I am looking to collaborate with AI safety researchers, engineers, and policymakers. Together I hope to build stress-testing environments for AI workflows, expand datasets to represent non-Western demographics, and help shape inclusive technology policy.
If these questions resonate with you, I would be glad to connect — please reach out via my Contact page.
Past Research & Industry Footprint
Before turning to AI safety, I spent several years building computer vision and industrial AI systems across both corporate and academic settings. During my MTech at NIT Durgapur, I developed a robust face-recognition system that fused complementary visual descriptors to strengthen cross-modal recognition.
As a Software Engineer at Sketchbytes Research Labs, I translated that theory into commercial products. I led the face-detection and biometric-authentication division and designed convolutional neural networks for high-precision medical imaging — including automated bone-fracture detection, disease detection, and body-region segmentation.
Moving into optical security research, I worked as a Computer Researcher (Holographer) at De La Rue in the UK. As part of a small research team, I engineered novel image-processing techniques to improve the security of holographic master plates for international clients — including Microsoft and the Central Bank of Azerbaijan — strengthening production quality and anti-counterfeiting measures.
Later, as Lead Data Scientist at Sortit AI in London, I built deep-learning solutions for the architectural sector. I constructed custom datasets and trained CNN models to identify structural elements from 2D floor plans, enabling automated building-cost estimation from a single drawing.