Professional Experience

Adobe Research — San Jose, CA

Research Scientist, Agentic Intelligence Research (AIR Lab) · Sept 2023 – Present
Manager: Tong Sun

  • Partial-policy gradients for RL in LLMs: Novel credit assignment technique for offline policy gradients that scales sub-horizon planning for multi-step LLM trajectories, with scaling laws for policy gradient finetuning and improved long-context persona-following in LLM roleplaying.
  • Reasoning-based LLM clustering via RL/GRPO: Post-trained Qwen models via GRPO to reason over complex user queries for organizing large text collections, improving query-aware clustering by +14% V-score (at par with GPT-o3).
  • Roleplaying agents: Multi-agent orchestration layer with planning, behavior alignment, and role verification for consistent specialist personas; RL reward modeling and curriculum learning for tutor agents.
  • Test-time agentic skill learning: Automatic skill discovery via experience-based distillation to build a hierarchical, self-evolving skill library, improving Claude Skills task success rate by 13%.
  • Self-evolving agents: Led development of a self-evolving tutor agent using TextAdam, a conversational prompt optimization technique yielding a 23% gain in instruction following over TextGrad and GEPA.
  • Agentic memory and continual learning: Novel memory architectures that let LLM agents learn from past interactions and reduce recurring errors; deployed in the Common Agentic Framework and used by 5+ Adobe products.
  • LLM attribution and RAG: Post-hoc attribution agents providing citations for QA systems over document text, tables, charts, and visual artifacts via multimodal retrieval and agentic verification; improved table attribution in Acrobat AI Assistant by 12%.

Microsoft Research — Redmond, WA

Research Scientist Intern, Visual Document Understanding · May 2023 – Aug 2023
Manager: Dinei Florencio

  • Developed a Transformer-based model for chart information extraction that recovers chart structure and infills underlying data values, improving F1 score by 15% over prior methods.

Meta AI (Facebook AI) — Menlo Park, CA

Research Scientist Intern, NLP & Speech Team · Sept 2022 – May 2023
Mentor: Zhe Liu

  • AI for speech recognition and language model personalization: retrieval-augmented language modeling for NLP+speech, generative NLP for factual consistency and domain adaptation, and embedding-based retrieval for query understanding, improving word-error rate in ASR across several datasets. Papers accepted to EMNLP 2023 and Interspeech 2023.

Adobe Research — San Jose, CA

Research Scientist Intern, Document Intelligence Lab · June 2022 – Aug 2022
Mentor: Vlad Morariu

  • Automated document editing based on natural language prompt instructions. Paper accepted to AAAI 2023.

Verisk AI — New Jersey, NJ

AI/ML Research Intern · Jan 2022 – June 2022
Mentor: Maneesh Singh

  • Natural language inference on legal documents using optimal evidence selection. Paper accepted to EMNLP 2022.

Adobe Research — San Jose, CA (Remote)

Research Scientist Intern, Document Intelligence Lab · June 2021 – Aug 2021
Mentor: Vlad Morariu

  • End-to-end extraction of hierarchical document structures from semi-structured forms and invoices. Paper accepted at WACV 2023.

Dataminr Inc. — New York, NY

AI Research Intern · June 2020 – Aug 2020
Mentors: Joel Tetreault & Alejandro Jaimes

  • Duplicate detection of news headlines as a tiered (event-entity-phrase level) paraphrase identification task, deployed as an internal system.

Education

University of Maryland, College Park — MD, USA
Ph.D. in Computer Science — AI, NLP and Multimodal Deep Learning · December 2023
Advisor: Dr. Dinesh Manocha

University of Maryland, College Park — MD, USA
M.S. in Computer Science, GPA: 3.97/4.0 · Aug 2019 – May 2021
Advisor: Dr. Dinesh Manocha

Netaji Subhas Institute of Technology (NSIT) — New Delhi, India
B.E. in Computer Engineering, GPA: 3.8/4.0 · Aug 2014 – May 2018

For talks, workshops organized, and reviewing service, see the Talks & Service page.