Hi, I’m Anirudh, a Research Associate at Adobe Research. I graduated with a Bachelor of Technology, majoring in Computer Science and Engineering with a minor in Robotics and Artificial Intelligence from the Indian Institute of Technology Guwahati (IIT Guwahati). I am deeply passionate about interpretability in foundation models and applying these insights to develop robust, efficient solutions for real-world applications.

Research

At Adobe Research, I collaborate with Dr. Apoorv Saxena and Dr. Balaji V. Srinivasan on advancing the attribution engine for Adobe Acrobat AI Assistant, a system that provides document references as evidence for AI-generated answers. Notably, I led the development of the technology powering attribution in contracts, which was highlighted by Adobe’s CEO in a Fox Business interview (6:45–7:15).

My current focus is on improving the attribution feature by incorporating multimodality and multilinguality. Additionally, I have also worked on detecting and mitigating textual/visual hallucinations, optimizing LLM inference, improving context focus in LLMs and revisualising infographics for better user consumption. I tend to approach these challenges by leveraging model internals and explainability to ensure transparent and reliable AI systems.

During my undergraduate studies at IIT Guwahati, I completed my bachelor’s thesis under the guidance of Prof. Arijit Sur. My research explored Visual Representation Learning, specifically harnessing inherent intra-class variation to improve contrastive learning methods. This approach provided richer and more effective contrasts compared to traditional data augmentations. Previously, I worked with Prof. Shyamanta M. Hazarika on grasp recognition, employing techniques like Inverse Reinforcement Learning and high-level semantic information such as object affordances.