curriculum-vitae

Graduate Research Assistant

Research Foundation for SUNY

Guided by Dr. David Doermann, A2IL lab.

Software Engineer

Veritas Technologies LLC

Developed a novel algorithm for predictive execution of resource-intensive tasks that reduced execution time by 56%. Demonstrated a paper at Veritas's annual technical conference, named IDLEBOT. Bestowed Certificate of Merit 2020. Developed Ransom-ware Detection tool modeling file change log in Veritas Filesystem (VxFS). Presented project in Veritas's annual Hackathon, 2021. Lead effort to start ML for Systems group in Org.

Engineering Intern

Veritas Technologies LLC

Reduced execution time of variants of ‘ls' command by 20% using directory inode read ahead algorithm.

Doctor of Philosophy

University at Buffalo, The State University of New York

Computer Science

Master of Science

University at Buffalo, The State University of New York

Computer Science

Bachelor of Technology

Walchand College of Engineering

Information Technology

Google Research Mentorship Program Scholar

Google

Accepted to a three-month program that matches students with Google mentors and peers to support their pursuit of computer science research pathways.

Certificate of Merit - Veritas Illuminate

Veritas Technologies LLC

Presented a paper named 'IDLEBOT - Predictive execution of resource-intensive tasks' in Veritas's annual technical conference.

World Code Sprint

Hackerrank

Received silver medal.

Honorable Mention ACM-ICPC 2015 Regionals

Association of Computing Machinery and Baylor University

Represented Walchand College of Engineering FIRST time in International Collegiate Programming Contest regionals at Amrita University, Coimbatore. It is sponsored by IBM and conducted by ACM in association with Baylor University.

Best project competition

Department of Information Technology, WCE, Sangli

Ranked 2nd best project in the department for calculating top 10 Twitter hashtags using Apache Spark, Beautiful Soup, and Twitter4j API for real-time processing and sentiment analysis, with results visualized using Javascript D3.

First Runner Up Code-junkie MINDSPARK-2K14

College of Engineering, Pune

An algorithmic coding competition with over 500 students in teams of 2 from different colleges in India. It had 3 rounds: MCQ, short coding challenge, and long coding challenge.

DST Inspire Science camp

Indian Institute of Information Technology, Pune

Selected (Top 1%) to participate in Science camp for a week hosted by the Department of Science and Technology, Government of India, and Indian Institute of Information Technology, Pune.

ChartReformer: Natural Language-Driven Chart Image Editing

International Conference on Document Analysis and Recognition

Pengyu Yan, Mahesh Bhosale, Jay Lal, Bikhyat Adhikari, David Doermann

LineFormer: Rethinking Line Chart Data Extraction as Instance Segmentation

International Conference on Document Analysis and Recognition

Jay Lal, Mahesh Bhosale, Aditya Mitkari, David Doermann

Unpaired Text and mask conditioned histopathology image synthesis with diffusion models

Using unpaired text reports and label masks to control histopathology image generation.

Action spotting

Using object specific features and their positional correspondence in a frame for temporally precise action detection.

Line-Former

Transformer based instance segmentation for document chart data extraction achieving SOTA results across many benchmark chart datasets. Submitted the paper to ICDAR 2023.

ROS

Implemented BUG2 obstacle avoidance, Camera calibration, Monocular VO, A* path planning, Bays filter for state estimation for ClearPath Husky in ROS.

Re-identification SoccerNetv3

Soccer player re-identification leveraging two-stream (RESNET + OpenPose sub-network) deep neural network using layer-wise triplet similarity loss. Employed Bilinear pooling to pool features from two streams. Beat SOTA OSNET by 2.1% in MAP and 1.8% in IOU.

IDLEBOT

Developed command line tool to recommend optimal time slot for execution of resource-intensive tasks for Veritas products using LSTM, FB-Prophet, and ARIMA. The proposed novel 'Weighted time-slot selection algorithm' to select an optimal slot achieved a 56% reduction in exec. time.