Neel Bhandari
Hello! I am a Masters student in the Language Technologies Institute within the School of Computer Science at Carnegie Mellon University. I am currently working with Prof. Maarten Sap on research that focuses on developing neurosymbolic methods to enhance model safety and mitigate reward hacking, as well as investigating how LLM agents demonstrate different forms of social intelligence in non-cooperative scenarios.
Gates Hillman Center
5000 Forbes Avenue
Pittsburgh, PA 15213
I am spent the past summer interning at Adobe as an ML Intern, where I developed agentic frameworks for creative content generation and performance marketing within Adobe GenStudio.
Prior to joining CMU, I spent two fantastic years at PayPal as a Machine Learning Scientist. There, I developed PayPal’s first automated root cause identification tool, which extracts root causes and problem statements from millions of customer chat transcripts. I was also part of the team that worked on the first LLM-based PayPal Assistant chatbot.
My research has focused on multilingual NLP and robustness. I contributed to the Aya Project, where we developed a state-of-the-art multilingual model covering 100 languages (We won Best Paper at ACL 2024!). My specific contributions included efficient data pruning and pipelining to create an accessible model. Additionally, I collaborated with Dr. Pin-Yu Chen at the MIT-IBM Watson AI Lab, studying the robustness of text-based adversarial attacks from a multilingual lens.
selected publications
- Best Paper at ACL 2024Aya model: An instruction finetuned open-access multilingual language modelarXiv preprint arXiv:2402.07827, 2024
- Out of Style: RAG’s Fragility to Linguistic Variation2025
- Lost in translation: Generating adversarial examples robust to round-trip translationIn ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023