Tushar Pawar

Tushar Pawar

ABOUT ME
Deep Learning practitioner, Algorithmist, Doer, Sr. Computer Vision Engineer @ BigApp Technologies
Deep Learning practitioner, Algorithmist, Doer, Sr. Computer Vision Engineer @ BigApp Technologies

Senior Computer Vision Engineer @ BigApp Technologies. I have in depth mathematical understanding of Machine Learning concepts being a Deep Learning practitioner for last 4 years. I am also a Competitive Programmer and regularly participate in online competitions on HackerRank, Codechef and also participated in ACM-ICPC 2016. Most of my work day consists of building deep learning based computer vision systems and deploying them to production.

English
Mumbai (+05:30)
Joined August 2018
EXPERTISE
3 years experience
My love for Computer Vision started a few years ago when I used to build image manipulation tools in Matlab. Started with toy problems su...
My love for Computer Vision started a few years ago when I used to build image manipulation tools in Matlab. Started with toy problems such as number plate detection in images of cars, face detection and image character recognition for number plates using template matching technique. Slowly the projects started to grow bigger and the problems required more than just a naive image processing approach. Thats when I moved to Python and OpenCV library to do much more complex manipulation to images. Over the years, I have built and deployed multiple computer vision based solutions which include deep learning based optical character recognition for transcription of scanned document images. This project consisted of image cleaning and noise removal, and a neural network built in Tensorflow to make character by character predictions. This project achieved higher accuracy that Tesseract OCR engine which is considered the best open-source library for the task. I am currently working on a facial recognition system that will be integrated with security surveillance systems to identify people for access management and threat detection.
6 years experience
I have been working on various projects written from sctratch in Python. I strive to not only write the code to get the job done, but als...
I have been working on various projects written from sctratch in Python. I strive to not only write the code to get the job done, but also to make it more readable and maintanable. Stritly follow PEP8 coding conventions. I am able to impliment complex algorithms in Python and write documentation for all the code I write for any project.
3 years experience
My daily work includes implimenting deep learning algorithms in Python, making end user applications and deploying them to cloud. These d...
My daily work includes implimenting deep learning algorithms in Python, making end user applications and deploying them to cloud. These deep learning solutions are fast, robust and scalable.
InceptionCnnLstm
View more
InceptionCnnLstmTensorFlowFace RecognitionComputer Vision
View more
6 years experience
All the projects that I work on is version controlled using the git tool. My expertise in git has helped me in fast prototyping of ideas ...
All the projects that I work on is version controlled using the git tool. My expertise in git has helped me in fast prototyping of ideas while development. It has also helped me manage several projects which are being developed in a team and use of git facilitated in team collaboration. Checkout my github profile at https://github.com/Backalla/ for more info on my projects
5 years experience
C++ was my first programming language which introduced me to the world of competitive programming. It was my language of choice while I ...
C++ was my first programming language which introduced me to the world of competitive programming. It was my language of choice while I used to solve complex mathematical algorithms and data structures problems on Hackerrank, SPOJ, UVA etc. Competed in ACM ICPC 2016 and also won many online competitive programming contests.
3 years experience
I have developed and deployed many deep learning models for end users using Amazon Web Services which includes S3 for data storage, EC2 f...
I have developed and deployed many deep learning models for end users using Amazon Web Services which includes S3 for data storage, EC2 for computation and web hosting and SQS for managing request and response queues.
Amazon EC2Amazon S3APILinuxAmazon SQSSoftware Deployment
View more
SOCIAL PRESENCE
GitHub
Dinho
A GRBL based plotter that can write (or draw) in my handwriting.
Python
2
0
Taz3D
DLP 3D printer on Raspberry pi.
JavaScript
1
2
Stack Overflow
51 Reputation
0
0
7
EMPLOYMENTS
Machine Learning Research Engineer
Preeminence
2017-06-01-2018-07-01
Responsibilities include designing and implementing complex architectures for projects and leading the research and development for execu...
Responsibilities include designing and implementing complex architectures for projects and leading the research and development for executing them. Developed an internal library on top of Tensorflow to speed up the process of development of machine learning models. This library reduces the amount of new code by 50-60%.
Python
OpenCV
TensorFlow
View more
Python
OpenCV
TensorFlow
View more
PROJECTS
Machine Learning based Multilingual Optical Character Recognition
Praemineo
2017
Developed a multilingual optical character recognition system which is able to extract text from the image of a document with 96% accurac...
Developed a multilingual optical character recognition system which is able to extract text from the image of a document with 96% accuracy using deep neural networks. My responsibilities in the development of this product includes - Detecting the presence of text in the image and isolating it by its coordinates using OpenCV library. - Designed the architecture to store this huge dataset (800K images) and getting it manually labelled. - Designed and implemented CNN+LSTM based model in tensorflow and trained it on the labelled dataset on AWS GPU machines. - The model achieved ~96% accuracy on the internal test data and ~93% on the external data in production. - Optimised this model to be deployed in production to deliver the OCR performance of ~1.5 seconds for an A4 page. - Currently working on scaling the system to recognise multiple languages and handwritten text.
Python
Deep Learning
TensorFlow
View more
Python
Deep Learning
TensorFlow
View more