Work Experience
- Software Developer, Finiq Consulting Pvt Ltd, Pune, India, July 2022 - Apr 2023
- Worked with 4 multinational banks including Bank of America
- Capabilities: Fix-to-fix connectivity in Java, developed Derivatives platform for BofA end-to-end
Academic Experience/Ongoing Projects
- Optimized Multi-Agent Debate (advised by Dr Xuan Wang and Dr Naren Ramakrishnan)
- In this work, we first show how sycophancy and cost affects the performance of multi-agent debate. To improve this, we introduce a new multi-agent debate framework. Using gradient descent and prompt optimization, we templatize the prompts that lead to the most accurate and efficient output for 7 reasoning datasets. With some additional metrics like confidence score and internal consistency score, our framework shows promising results in improving accuracy and efficiency for reasoning tasks. (Arxiv coming soon, Github coming soon, to be submitted to ACL 2025)
- ArguMentor: Augmenting User Experiences with Counter-Perspectives (advised by Dr Kurt Luther)
- As part of Advanced Human-Computer Interaction course
- Abstract: ArguMentor is a human-AI collaboration system that highlights claims in opinion pieces, identifies counter-arguments for them using a LLM, and generates a context-based summary of based on current events. It further enhances user understanding through additional features like a Q&A bot (that answers user questions pertaining to the text), DebateMe (an agent that users can argue any side of the piece with) and highlighting (where users can highlight a word or passage to get its definition or context). Our evaluation on news op-eds shows that participants can generate more arguments and counter-arguments and display higher critical thinking skills after engaging with the system. Further discussion highlights a more general need for this kind of a system.
- Ongoing work: Updated User study with Prolific, to be submitted to a conference in 2025
- Exploring Tailoring Bias in LLMs (advised by Dr Heng Ji, collaborators: Qingyun Wang, Chenkai Sun, Yi Fung)
- Summer Internship 2024
- Explored the bias LLMs display when asked to tailor their responses to specific demographics. 47 personas chosen, content, tone studied.
- Arxiv and github coming soon
- MultiAgent Adversarial Learning for Feedback Generation (advised by Dr. Henning Wachsmuth, collaborators: Omkar Joshi, Timon Ziegenbein)
- External collaboration
- Multi-agent framework to judge and give feedback to subjective debates
- Arxiv and github coming soon
- DebGraph: Graph Based Debate Evaluation and Feedback
- As part of Advanced Topics in Data (Learning with Graphs) class
- Graph-based system that combines the structural representation of Knowledge Graphs (KGs) with the contextual reasoning capabilities of Large Language Models (LLMs) to deliver multidimensional debate evaluation and feedback
- Github
- Evaluating Ghost Work in AI
- As part of Ethics in Computer Science course
- Reflection on ghost work in AI (inspired by Ghost Work by Gray and Suri), its ethical dilemmas, and how HCI could be a potential solution to it
