Research papers so far
Collection of research papers I have published so far
Research is the foundation of innovation. Over the past year, I've focused on exploring Machine Learning applications in document analysis and automated systems. Below is a detailed view of my published works and ongoing research.
1. Document Summarizer: A Machine Learning Approach to PDF Summarization
Published in: Proceedings of the 2025 International Conference on Smart Information and Artificial Intelligence Model Learning (ICSIAIML-25)
Publisher: Atlantis Press (Part of Springer Nature)
Authors: Piyush Dhoka, Sandip Thite, Varun Inamdar, et al.
DOI: 10.2991/978-94-6463-633-8_25
Abstract: This research presents a novel approach to PDF summarization using advanced Machine Learning techniques. The paper explores the integration of transformer-based models to extract key information and generate concise summaries from complex academic documents, significantly reducing the cognitive load for researchers.
Read Full Paper on Atlantis Press →
2. AttendifyAI: Research and Review Paper
Type: Review and Structural Research
Focus: Computer Vision & Attendance Systems
Overview
This review paper delves into the development of AttendifyAI, an AI-powered attendance tracking system. It covers the comparative analysis of various facial recognition algorithms and the architecture required for real-time processing in educational settings.
Key Contributions:
- Comparative study of facial recognition models.
- System architecture for scalable attendance tracking.
- Analysis of lighting and occlusion challenges in real-world environments.
3. Cost-Effective Edge AI Patent Concept: Crontab-Managed YOLOv8 Crowd Counter
Type: Ongoing Patent Concept
Focus: Edge AI & Public Safety
Overview
This paper proposes a low-cost Edge AI patent concept for automated crowd counting. The system integrates a crontab-controlled execution pipeline and a customized YOLOv8-based detector to improve public safety in crowded areas. It utilizes a microcontroller with a low-cost USB camera for on-device head tracking without network connectivity, ensuring visual privacy, minimizing operational costs, and eliminating network latency.
Key Contributions:
- Offline edge computing approach for crowd counting and tracking.
- Crontab-managed execution pipeline for YOLOv8.
- Validated via field testing with over 20,000 daily visitors during the Navratri festival at Shri MahaLaxmi Temple in Pune, India.
Future Work
I am currently working on scaling these models for larger datasets and exploring vision-language models for more intuitive document interaction. If you are interested in collaborating on research, feel free to reach out via my socials!
