Shravan Chaudhari

News

• Aug 2026SpIDER, on spatially informed dense embedding retrieval for software issue localization, is under review at EMNLP 2026. [arXiv]
• Jun 2026Paper accepted to ECCV 2026: Shared LoRA Subspaces for Almost Strict Continual Learning, with Prakhar Kaushik and Profs. Chellappa and Yuille. [arXiv]
• May 2026New preprint FLAME, on adaptive mixture-of-experts for continual multimodal multi-task learning, co-led with X. Han, with Profs. Chellappa and Saria — under review at NeurIPS 2026. [arXiv]
• May 2026Paper accepted to TMLR: Open-Set Domain Adaptation Under Background Distribution Shift, with Yoav Wald and Prof. Saria.
• Jan 2026Started as a Student Researcher with the GitHub Copilot Science team at Microsoft.
• Dec 2025Three preprints out: Open-Set Domain Adaptation Under Background Distribution Shift, The Universal Weight Subspace Hypothesis, and SpIDER for issue localization in code graphs.
• Sep 2025Wrapped up an Applied Scientist II internship with the Amazon Coding Agents team.
• Jun 2025EigenLoRAx presented at the CVPR 2025 Workshops.
• Apr 2025Paper accepted to TMLR: rethinking time encoders for dynamic graph learning.
• May 2023Joined the JHU CS PhD program, advised by Prof. Rama Chellappa and Prof. Suchi Saria.

Research interests

Trustworthiness and fairness in AI · LLM agents and code understanding · Multimodal and embodied agents · Graph-informed retrieval-augmented generation · Domain adaptation under distribution shift · Out-of-distribution and novel-category detection.

Publications

* denotes equal contribution. For a full list, see my Google Scholar.

Experience

• Jan 2026 – May 2026
  Student Researcher, Microsoft
  CoreAI · GitHub Copilot Science team
  GRPO-based RL fine-tuning of Qwen2.5-Coder and graph-aware scaffoldings for code repository navigation with cost-efficient LLMs; +13% file-level Recall@1 on the 14B model and +40% on the 7B.
• May 2025 – Sep 2025
  Applied Scientist II Intern, Amazon
  Coding Agents team
  LLM-driven dense embedding retrieval and code-graph navigation for GitHub-style issue localization across Python, Java, JS, TS.
• Aug 2024 – Dec 2024
  Applied Scientist II Intern, Amazon
  Search Experience team
  Multimodal LLM feedback and a novel metric for measuring search-page impact, validated against human annotators.
• May 2022 – Dec 2022
  ML & Software Co-Op, Johnson & Johnson MedTech
  Algorithms & AI team
  Semi-supervised active learning for surgical-robot video curation; cut annotation cost by over 70% while improving tool detection.
• Jan 2023 – May 2023
  Research Assistant, NYU Center for Data Science
  Advisor: Prof. Jacopo Cirrone
  Self-supervised representation learning for histopathology under unknown distributions.
• May 2021 – Oct 2021
  NSF / IRIS-HEP Fellow, Princeton Institute of Computational Science & Engineering
  Graph neural networks for tau-particle identification; lifted ROC-AUC from 0.70 to 0.85.
• Sep 2020 – May 2021
  Student Researcher, CERN (Undergraduate Thesis)
  Deep CNNs for high-energy particle classification with raw LHC detector images.

Education

• 2023 – Present
  PhD, Computer Science, Johns Hopkins University
  Advisors: Prof. Rama Chellappa, Prof. Suchi Saria · GPA 4.0/4.0
• 2021 – 2023
  MS, Computer Science, New York University (Courant)
  CGPA 3.83/4.0
• 2017 – 2021
  BE, Electronics & Instrumentation, BITS Pilani
  GPA 8.52/10