Experience

Selected research, industry, and open-source work. Roles and affiliations are stated explicitly.

Industry & Research

  • June 2026 - Sept. 2026
    Applied Scientist Intern
    Microsoft
    • Working on multimodal reasoning for LLM agents in Microsoft 365 Copilot.
  • June 2025 - Sept. 2025
    Applied Scientist Intern
    Amazon
    Supervised by Adam Gabryś and Ashwin Chandramouli
    • Worked on memory and personalization for Amazon Nova language models.
    • Developed LikeBench, a benchmark for evaluating subjective likability in personalized LLM responses.
  • Sept. 2024 - June 2026
    Graduate Researcher
    Vision Research Lab, UC Santa Barbara
    Supervised by Professor B. S. Manjunath
    • Studied efficient multimodal reasoning agents using self-play reinforcement learning.
    • Developed transformer-based methods for multi-month trajectory anomaly detection and temporally grounded video understanding.
  • Jan. 2024 - Sept. 2024
    Contractor
    Google (Remote)
    Worked with Martin Görner, Product Manager for Keras and TensorFlow
    • Worked on mathematical reasoning, AI-generated text detection, prompt recovery, and human-preference prediction for LLM responses.
  • July 2023 - June 2024
    Research Assistant
    Institute of Robotics & Automation, BUET
    Supervised by Dr. Shaikh Anowarul Fattah
    • Built SONICS, a 97K-song benchmark for detecting fully synthetic music from systems including Suno and Udio.
    • Designed an efficient long-audio architecture that improved F1 while reducing runtime and memory use; accepted at ICLR 2025.

Earlier Research

  • Feb. 2022 - May 2023
    Undergraduate Thesis
    Department of EEE, BUET
    Supervised by Dr. Shaikh Anowarul Fattah
    • Developed a semi-supervised transformer for joint semantic segmentation and depth estimation, published at WACV 2024.
    • Proposed local-global window transformers for monocular depth estimation, published in IEEE Sensors Journal.
    • Introduced ArtiFact, a 2.5M-image benchmark for detecting synthetic images from seen and unseen generators, published at ICIP 2023.
  • Feb. 2022 - May 2023
    Research Collaboration
    University of Texas at Dallas (Remote)
    Supervised by Dr. Mohammad Saquib
    • Developed a semi-supervised ensemble for attributing synthetic speech to known and unknown generation methods under perturbations.
  • Apr. 2020 - Aug. 2020
    Research Collaboration
    Princeton University (Remote)
    Supervised by Professor Sun-Yuan Kung
    • Implemented and evaluated efficient methods for COVID-19 lesion segmentation from CT scans, resulting in an IEEE Transactions on Artificial Intelligence paper.

Open Source & Community

  • Nov. 2019 - Present
    Kaggle Grandmaster
    Kaggle
    • Reached a best rank of 5th among 61,000+ competitors in the Code category.
    • Earned 48 gold and 14 silver Code medals, plus one gold and three silver Competition medals.
  • Aug. 2021 - Sept. 2023
    Developer Expert
    Weights & Biases Developer Expert program (Remote)
    • Tested experiment-tracking workflows, reported issues, and created tutorials for machine-learning practitioners.
    • Integrated Weights & Biases into YOLOv5 workflows for underwater object detection.
  • 2019 - Present
    Author / Contributor
    Open-source machine learning
    • Contributed multi-backend layers, utilities, and examples to Keras and TensorFlow workflows.
    • Resolved dataset and checkpointing issues in Hugging Face Datasets and YOLOv5.
    • Released reproducible TensorFlow implementations of GCViT, TransUNet, Conformer, and ContextNet.