Bias Detection Systems in Indian Context
- Part of a research team working on bias detection systems in Indian context under Prof. Balaraman Ravindran and Dr. Gokul S Krishnan.
- Developed novel evaluation benchmark using contrastive learning-trained encoder model to detect biases in LLMs.
- Created India’s first comprehensive bias dataset containing 2,575 sentences using AI-in-the-loop curation across bias types.
- Formulated loss functions for contrastive learning of encoder-based models for better demographic representations.
- Achieved an F1 score of 0.73 for bias detection and engineered evaluation pipelines for Open and Closed-Source LLMs.
- Trained and implemented LoRA Adapters for debiasing models in Indian Biases.
- Co-authored a research paper on bias detection systems submitted to Eighth AAAI/ACM Conference on AI, Ethics, and Society.
