Hello! My name is Ishaan Watts. I am a graduate from IIT Delhi and have completed my B. Tech. in Engineering Physics with a minor in Computer Science. Currently, I am working as a Pre-Doctoral Researcher at Google DeepMind under the guidance of Dr. Partha Talukdar. My research centers around continual learning in Large Language Models (LLMs), through model composition, to make their architectures modular. I am also interested in improving performance of LLMs for low-resource languages to promote fairness and accesibility of AI.

Prior to this, I was a Research Intern at Microsoft Research, where I worked with Dr. Sunayana Sitaram on evaluating multilingual capabilities of LLMs. My projects focused on benchmarking these capabilities [NAACL'24], exploring efficient fine-tuning techniques to enhance them [ACL'24], and developing an ecosystem to fairly evaluate the rising number of Indic LLMs [EMNLP'24]. Additionally, I collaborated with Dr. Akshay Nambi and Dr. Tanuja Ganu on Shiksha Copilot, an educational AI assistant released by Microsoft. I also worked with Dr. Adrian de Wynter from Microsoft Redmond in creating a toxic multilingual benchmark aimed at facilitating the safer deployment of LLMs [AAAI'25].

Keywords: Large Language Models, Multilinugality, Evaluations, Model Composition

In the past, I interned at Griffith University under Dr. Saiful Islam, focusing on malware detection using machine learning. At Udaan, I developed a user embedding framework using GNNs that descreased Udaan dispute loss by 2.45% through improved fraud detection. At Torch Investment Management, I worked on stock price prediction models and integrated NLP for market sentiment analysis. For my bachelor's thesis at IIT Delhi, I collaborated with Prof. Abhishek Iyer on particle physics projects, utilizing machine learning for anomaly detection and particle identification.

For more details about my background, refer to my CV. If you'd like to chat with me about my work or research in general, feel free to reach out via email!

Last Updated: 30/1/2025 January, 2025: We have released the cultural artefacts collected in our work "PARIKSHA". [Link]
December, 2024: Our work "RTP-LX: Can LLMs Evaluate Toxicity in Multilingual Scenarios?" got accepted in the AI for Social Impact track at AAAI 2025!
November, 2024: I will be attending EMNLP 2024 in Miami, Florida to present our work.
September, 2024: Our work "PARIKSHA: A Large-Scale Investigation of Human-LLM Evaluator Agreement on Multilingual and Multi-Cultural Data" got accepted at EMNLP 2024!.
September, 2024: Gave an invited talk, titled "Evaluations in Multilingual and Multicultural World", at the National Research Council Canada (NRCC).
August, 2024: I will be attending ACL 2024 in Bangkok, Thailand to present our work, "MAPLE: Multilingual Evaluation of Parameter Efficient Finetuning of Large Language Models".
July, 2024: I have joined as a Pre-Doctoral Researcher at Google DeepMind under Dr. Partha Talukdar!

C=Conference, ^*=Equal Contribution