As a Senior Machine Learning Engineer at Bosch in Coimbatore, I am engaged in the development of advanced AI applications, including Advanced Driver-Assistance Systems (ADAS) and the creation of a foundational language model that facilitates the conversion of manual test cases into automated test scripts using straightforward English instructions. I hold a Master of Technology in Computer Science from Jadavpur University, Kolkata, where Prof. Dr. Kamal Sarkar guided me. My research is centred on Causal AI, Deep Learning Theory, and LLM Interpretability specifically focusing on understanding the internal mechanisms of large language models and leveraging the interpretability findings to enhance the performance of downstream tasks. I believe mechanistic interpretability is the key to AGI.

Updates

• ✅ [COLING-2025(Main Track)! Acceptance] CryptOpiQA: A new Opinion and Question Answering dataset on Cryptocurrency

• ✅ [COLING-2025(Industry Track)! Acceptance] On the effective transfer of knowledge from English to Hindi Wikipedia

• 🚩 Check Out our New Preprint on Bias Mitigation - Do the Right Thing, Just Debias! Multi-Category Bias Mitigation Using LLMs

• ✅ [TMLR! Acceptance] Size Lowerbounds for Deep Operator Networks 💎

• Currently working at Bosch with ADAS program team, with a focus on Driver Monitoring and Assistance space.

• My work on Bengali Text Classification, which introduces a new multi-class dataset and evaluates various machine learning and deep learning models, is currently under review for publication in Language Resources and Evaluation.

• Our team’s project on creating a foundatational language “text to robotframework script generation” model is nearing completion with promising results for automation testing enhancements.

Publications

• Under Review
  • “Bengali Text Classification: A New multi-class Dataset and Performance Evaluation of Machine Learning and Deep Learning Models”, Language Resources and Evaluation, 2024. Read Preprint
  • Bias Mitigation Using Reinforcment Learning, ARR, 2024
• Journal
  • A. Mukherjee, A. Roy, “Size Lowerbounds for Deep Operator Networks”, Transactions of Machine Learning Research (accepted), 2024.
• Prepint
  • Amartya Roy, Danush Khanna, Devanshu Mahapatra, Vasanthakumar, Avirup Das, Kripabandhu Ghosh, “Do the Right Thing, Just Debias! Multi-Category Bias Mitigation Using LLMs”, arxiv link - September 24, 2024
• Workshop
  • Paramita Das, Amartya Roy, Animesh Mukherjee, “WikiTransfer: Knowledge Transfer from High-Resource to Low-Resource Language in Multilingual Wikipedia”, Wiki Workshop (11th edition) – June 20, 2024.

Education

• Master of Technology in Computer Science, Jadavpur University, Kolkata, India, 2020-2023, CGPA: 8.76
  • Advisor: Prof. Dr. Kamal Sarkar
• Bachelor of Technology in Electronics & Electrical Engineering, Kalinga Institute of Industrial Technology, Bhubaneswar, India, 2016-2020, CGPA: 8.82
  • Advisor: Prof. Rishi Khanna

Achievements

• Selected for ACM India Summer Schools, 2022.
• Selected for Indian Statistical Institute Winter Deep Learning School, 2023.