I am a Research Fellow at MATS and work with Proximal on AI4Code, Formal Methods, reward hacking, benchmarking and evals.

My research focuses on how AI systems can write correct and verifiable code, and reason with human-like flexibility. I work at the intersection of AI4Code, AI4Math, Formal Methods, and Human-AI Interaction. Recently, I developed SpecGuard, which uses Lean 4 to detect and formally certify conflicts between coding-task descriptions and tests before an agent acts.

Previously, I worked at Microsoft Research’s PROSE team on autonomous coding agents, conversational debugging, and rubric-based evaluation, with contributions deployed in Visual Studio and GitHub Copilot. A recurring thread of interest in my research is translating informal intent into machine-checkable specifications. I started with mathematical autoformalization in Lean 4, then into verified code generation and, at Qualcomm, autoformalizing natural-language requirements for semiconductor design verification.

Outside of research, I like to swim, dive, read manga, and play grand strategy video games.

Check out my CV, or reach me through email.

Publications

SpecGuard: Proving a Task Is Broken Before the Agent Cheats
Param Biyani, Krishnamurthy Dvijotham
Preprint 2026
pdf | code

ShutdownMod: Turning Any Trajectory into a Shutdown Evaluation
Param Biyani, Divij Chawla, Elliott Thornley, Christos Ziakas
Interpreting Agent Behavior at NeurIPS 2026
pdf

Improving Language Agents Through BREW
Shashank Kirtania, Param Biyani, Priyanshu Gupta, Yasharth Bajpai, Roshni Iyer, Sumit Gulwani, Gustavo Soares
COLM 2026 | Multi-Turn Interactions in LLMs at NeurIPS 2025
pdf

IndiMathBench: Autoformalizing Mathematical Reasoning Problems with a Human Touch
Param Biyani, Shashank Kirtania, Yasharth Bajpai, Sumit Gulwani, Ashish Tiwari
Post-AI Formal Methods at AAAI 2026
pdf | code

🏆 Best Paper Presentation Award
RUBICON: Rubric-Based Evaluation of Domain-Specific Human AI Conversations
Param Biyani, Yasharth Bajpai, Arjun Radhakrishna, Gustavo Soares, Sumit Gulwani
AIware 2024
blog | pdf

🏆 Best Paper Award
Let’s Fix this Together: Conversational Debugging with GitHub Copilot
Yasharth Bajpai, Bhavya Chopra, Param Biyani, Cagri Aslan, Sumit Gulwani, Dustin Coleman, Chris Parnin, Arjun Radhakrishna, Gustavo Soares
VL/HCC 2024
pdf

Exploring Interaction Patterns for Debugging: Enhancing Conversational Capabilities of AI-assistants
Bhavya Chopra, Yasharth Bajpai, Param Biyani, Gustavo Soares, Arjun Radhakrishna, Chris Parnin, and Sumit Gulwani
HCI-NLP NAACL 2024
pdf