About

I am a fifth-year PhD student in computer science at the University of Chicago, advised by Dr. Allyson Ettinger. I study how LLMs can learn over time and how LLM-based agents can work together.

My research spans data selection, continual learning, and multi-agent communication and evaluation. My work examines whether training-data coverage matters more than data difficulty, how to improve a target capability while limiting unintended changes to model behavior, and how access to information affects coordination among agents. I also work closely with Dr. Claire Donnat and Dr. Kanishka Misra.

Before my PhD, I worked in industry on distributed file systems and data replication. I did my M.Tech. in Computer Science at IIT Kanpur.

Feel free to reach out at lcpandia@uchicago.edu with questions, ideas, or research you’re excited about.

Publications

  • SAFE-DRIFT: Data Selection for Supervised Fine-tuning with Controllable Off-Target Drifts Yeo Jin Jung, Yating Liu, Lalchand Pandia, Claire Donnat NeurIPS 2026PDF
  • Rethinking Data Selection: The Importance of Coverage over Difficulty in Generative Fine-Tuning Lalchand Pandia, Kanishka Misra, Allyson Ettinger DATA-FM Workshop, ICLR 2026PDF
  • Sorting Through the Noise: Testing Robustness of Information Processing in Pre-trained Language Models Lalchand Pandia, Allyson Ettinger EMNLP 2021PDF
  • Pragmatic Competence of Pre-trained Language Models through the Lens of Discourse Connectives Lalchand Pandia, Yan Cong, Allyson Ettinger CoNLL 2021PDF

Industry experience

  • System ArchitectDec 2018 – Jul 2022
    Huawei Technologies
    • Designed the replication module for the Huawei file system.
    • Designed a distributed data manager for cross-cluster storage replication.
  • Member of Technical Staff IIJul 2017 – Jul 2018
    NetApp
    • Built data-backup functionality for moving data from storage servers to backup devices.
  • System EngineerDec 2012 – Jun 2015
    Tata Consultancy Services
    • Built Java client-server applications for insurance policy management.