Publications

Working Papers & Submissions

1. Nexus Companion: Emotion-Labeled Dataset for Empathetic LLMs (EMNLP Submission)
   Lead researcher for NexusCompanion, a scalable benchmark for multi-label emotion classification in LLMs. Curated 100K Reddit conversations labeled using Plutchik's Emotion Model for emotional alignment. Fine-tuned models with Hugging Face, LoRA, and GPT-4; optimized for inference efficiency and fairness. Collaborating with Dr. Sumon Biswas and Dr. Max Han. Submitted to EMNLP 2025. </br>
2. Advancements in Counterfactual Explainability with GNNs
   Developing dynamic graph neural network methods for counterfactual explanation in LLMs and deep models. Focus on symbolic interpretability, static analysis, and causal graph modeling. Advisor: Dr. Jing Ma.
3. Region-Specific LLMs with Fairness-Aware LORA Optimization
   Designed fine-tuning techniques for bias mitigation in LLMs across dialects and regional language variants. Benchmarked fairness with contrastive prompts and symbolic consistency checks.
4. Trustworthy AI and Formal Fairness Research
   Contributed to formal fairness testing pipelines for ML models. Surveyed 75+ papers across symbolic, formal, and empirical fairness auditing frameworks.