Hi, I’m Rishi!

I am a master’s student at the University of Maryland, studying data science with a focus on building reliable, interpretable, and impactful machine learning systems.

Contact:
Domains
Machine Learning Deep Learning Computer Vision Natural Language Processing Transformers PyTorch C SQL LLMs Vision-Language Models Tesseract OCR Docker Pandas NumPy Matplotlib OpenCV Scikit-Learn
Experience

Bachelor’s Thesis Student

Jan 2026 - Apr 2026

Visual Data Computing Group (VisDom) - IISER Bhopal, India - On-site

Grade: A

Title: Query-Aware Token Budgeting for Efficient Late-Interaction Visual Document Retrieval

  • Proposed a two-stage, query-aware token selection technique to address the high storage and query-time cost of late-interaction visual document retrievers.
  • Demonstrated efficient late-interaction via query-specific token selection across 10 ViDoRe datasets in ColMordenVBERT; achieved 98.39% PF1 recovery at a PF8-equivalent budget via greedy selection, while token top-k method established the practical latency-quality frontier.

Data Science Intern

May 2025 - Jul 2025

SpectraX Management LLC - Houston, TX, United States - On-site

  • Built a machine learning based document classifier for bulk classification of distorted and scanned documents.
  • Implemented robust preprocessing techniques, utilized Tesseract for OCR and Python libraries such as OpenCV and Scikit-Learn, and deployed the backend tool.
  • Developed adaptive thresholding and deskewing heuristics to handle extreme scan distortions, elevating classification precision to 0.91 and accelerating bulk processing speeds by 75.4%.
Publications

R. P. Sivakumar, R. R. Dwivedi, and V. K. Kurmi, "Query-Aware Token Budgeting for Efficient Late-Interaction Visual Document Retrieval", 2026 IEEE International Conference on Data Mining (ICDM), 2026. [Accepted]

https://arxiv.org/abs/2609.07262