Abhijit Balaji PDF
Abhijit Balaji PDF
Abhijit Balaji PDF
edu
Website : abhijit-2592.github.io Mobile : +1-214-603-5538
GitHub : github.com/Abhijit-2592 LinkedIn : linkedin.com/in/abhijit-balaji
Education
The University of Texas at Dallas. Dallas, Texas
•
Master of Science in Computer Science; GPA: 3.66/4.0 Aug. 2019 – Jun. 2021
VIT University Vellore, India
•
Bachelor of Technology in Mechanical Engineering; CGPA: 9.01/10 May. 2013 – May. 2017
Work Experience
Explosion Berlin, Germany
•
Research Assistant (Internship) Dec 2019 - Jan 2020
As part of my internship, I was maintaining and improving multiple high impact in-house and external open source
projects. Worked with the creators of SpaCy in developing a new machine learning library called Thinc. Also, improved the
performance of a Neural Network based Coreference Resolution system created by the HuggingFace team.
SPi Global Chennai, India
•
Data Scientist Jul 2018 - Jun 2019
Identified potential business problems in the e-publishing industry and played a crucial role in semi-automating a lot of
manual processes using AI thereby, improving the productivity of the company. Given below are a select few of my projects
at SPi-Global.
◦ Chart-Text: An AI-based system that generates textual description of a given chart image. The project provides a useful
aid to blind people in understanding chart images in the digital content. It uses CNNs and other conventional image
processing algorithms to accomplish the same.
◦ Medical Journal Indexing: An extreme multi-label classification for medical journal indexing involving 100,000 labels
using deep neural networks. This project improved the productivity of human indexers by over 3 times by suggesting them
with suitable indices out of the 100,000 options for medical journals.
◦ Robo-Reader: An AI-based proof-reading and copy-editing system for books, journals, and documents. This project
improved the productivity of human copy-editors by over 5 times by helping them to focus more on the places where copy
editing is required in the document.
Endotherm Fluids India Pvt LTD Chennai, India
•
Data Science Engineer May 2017 - Jun 2018
◦ AI based Machine Vision System: Created an AI-based Visual Inspection System that detects and classifies surface
defects in metal components. The system uses CNNs to accomplish this. The system effectively replaced 10 skilled quality
inspectors in the company and resulted in twice as efficient productivity and a significant decrease in customer complaints.
This project also paved the way for a new startup company ”JIDOKA Technologies” based in Chennai India.
Personal Projects
• Active learning workflow with Tensorflow object detection API and Prodigy - an image labeling tool from the
makers of spaCy: Worked with the creators of spaCy(explosion.ai) to create an active learning pipeline using Tensorflow Object
Detection API for their labelling tool Prodigy
• Keras object detection: Faster-RCNN architecture re-implementation in Keras. This project attempts to reproduce the official
Faster-RCNN paper. Additionally, this can be used to convert any image classification model trained in Keras into an object
detection model.
• model-server: A fully flexible python based gRPC server for hosting Deep Learning and Machine Learning models trained on any
framework.
Programming Skills
• Languages: Python, C/C++, Java, Matlab
• Machine Learning Libraries: TensorFlow, Keras, PyTorch, SpaCy, Thinc, Scikit-Learn
• Technologies: Docker, MongoDB, Git
• Design Softwares: Solidworks, ANSYS
Research Publications
• Chart-Text: A Fully Automated Chart Image Descriptor
Invited Talks
• Visualizing and understanding what CNNs learn: This talk was given at SRM Institute of Science and Technology -
(Chennai) as part of their Faculty development program