Hi! I am Shashank Kirtania (he/him), pre-doctoral research fellow at Microsoft PROSE team. I collaborate with Gustavo Soares, Arjun Radhakrishna and Sumit Gulwani. My current research is focused towards use of prompts as programs, and reliable code generation. I am highly interested in mechanistic interpretability of LLMs, Automated Theorem Proving and Program Verification.
I am applying for PhD positions in the Fall'25 cycle.
Previously, I have been privileged to work with Soma Dhavala and Makarand Tapaswi during my time at Wadhwani AI , developing an AI4Education solution for Indic languages that was deployed in the Indian state of Gujarat. This initiative has completed over 1,000,000 successful assessments in the first 6 months of deployment.
I have also worked closely with Thomas Wiecki at PyMC Labs, where I spent most of my time improving the workflow of Bayesian models present in PyMC.
My more formal curriculum vitae could be found here.
-
STACKFEED: Structured Textual Actor-Critic Knowledge Base Editing with Feedback preprint
Shashank Kirtania*, Naman Gupta*, Priyanshu Gupta, Krishna Kariya, Sumit Gulwani, Arun Iyer, Suresh Parthasarathy, Arjun Radhakrishna, Sriram K. Rajamani, Gustavo Soares
Under review
-
MetaReflection: Learning Instructions for Language Agents using Past Reflections arxiv
Shashank Kirtania*, Priyanshu Gupta*, Annanya Singha*, Sumit Gulwani, Arjun Radhakrishna, Sherry Shi, Gustavo Soares
Accepted at EMNLP’24 Main * = shared first author
-
LOGIC-LM++: Multi-Step Refinement for Symbolic Formulations arxiv
Shashank Kirtania, Priyanshu Gupta, Arjun Radhakrishna
Workshop on Natural Language Reasoning and Structured Explanations @ ACL 2024
-
DWT-CompCNN: Deep Image Classification Network for High Throughput JPEG 2000 Compressed Documents link
Tejasvee Bisen, Mohammed Javed, Shashank Kirtania, P Nagabhushan
Pattern Analysis and Applications, Springer, 2023