Rahul Singh
I am a Data Scientist based out of New Delhi, India. My primary interests include Data Science,Big Data,Data Counsultant, Machine Learning, Deep Learning,Natural Langauge Processing,Statistical Modeling,Business Intelligence and Data Analytics.
I have focused on deriving useful business insights from large datasets and helping organizations generate business value and revenue using an agile and data-driven approach.
I obtained my Master's Degree MCA(Master of Computer Application)from ABES Engineering College, India in 2009.
I obtained my Bachelor's Degree(B.Sc) in Computer Application from Chhatrapati Shahu Ji Maharaj University, India in 2005.
Prior to joining Kochar, I've had the pleasure of working at:
Email /
CV /
LinkedIn /
Github
|
|
Research Interests
My interests broadly lie in the fields of Data Science, machine learning, and deep learning and Big Data.
Much of current work research is focused on supervised learning, unsupervised learning, and dataless classification i.e. training machine learning models with little or no data.
|
|
Road Network Extraction using Satellite Imagery
Submission for MoveHack Global Mobility Hackathon 2018
Github /
Demo video
The project involved training and deploying a road segmentation and extraction system using high-resolution satellite imagery for reliable and low-cost terrain monitoring and infrastructure quality assessment.
Optimized the trained model for real-time applications with final inference speed of only 0.28 seconds on Tesla K80 GPU achieving a mask accuracy of 95% and a dice score of 65% on the validation set.
|
|
LingoLens
@CampK12 with Anshul Bhagi
Slides /
API Docs
A multilingual language learning app for K12 students providing translations and transliterations for indoor and outdoor objects in over 20 languages.
We implemented a real-time object detection and on-device scene classification model Trained a custom YOLO model to detect ∼150 objects with an mAP of 67% improving the on-device detection and classification inference speeds by 19% with only 6% performance degradation compared to SoTA methods such as RetinaNet and SSD.
|
|
Book Genre Classification
@Spikeway Technologies with Praveen Kumar
Github /
Demo
In this project, I led a 3 member team a ML-based system to classify books into their genres purely based on its title, without prior knowledge or context of author and origin.
I was responsible for the architecture, training, and deployment of the model. We improved the TFIDF and LR baselines by training a LSTM using pre-trained word2vec embeddings.
|
|
Computer Vision Projects
Just for Fun
Github
Open-source implementations for various computer vision tasks such as Template matching, Object Tracking, Face Swapper, Live Sketch, etc, using OpenCV.
60+ stars, 60+ forks.
|
|
TravelCamp
Submission for IMAD 2016
Github /
Demo
This project involved developing a social blog/profile web app where users can interact with each other. Other features including personal feed, posts, comments, signing, and logging in.
The web app follows RESTful approach for CRUD operations and was deployed using Heroku. Backend frameworks used: NodeJS, ExpressJS, MongoDB, PassportJS
|
|