- Collected 5,38,000 satellite images of Bengaluru. Processed all images to find the percent of green using OpenCV.
- Performed reverse geo-coding to map area name to co-ordinates.
- Implemented a custom multi-threaded map-reduce in python.
|
- Implemented a n-layer neural network from scratch using pure Julia.
- Supports binary classification.
- Support for training on GPUs using CuArrays.jl, CUDAnative.jl. Optimised using custom CUDA kernels.
|
- Implemented Truncated SVD in training fully connected DNNs.
- Improved training speeds by 1.2 to 3 times along with increased accuracy.
- Experimented with different values of (new) hyper-parameters (more details in the report).
|
- Made a react native app with integrations for Firebase - login, storage, database, cloud messaging, Tensorflow Lite object detection (for detecting vehicles in the picture and its number plate) and API calls to the server.
- Built multiple APIs in Flask for Number plate recognition (OCR) and Geospatial indexing which is used to notify nearby users.
- Tools used: Python, React Native, Flask, Firebase, Tensorflow, TFLite, Geopandas.
|
- Created a custom dataset named CapStyle5k derived from Flickr8k.
- Implemented a sequence-to-sequence encoder decoder model with attention.
- Tools used: Python, Keras, Tensorflow GPU, CUDA, Flickr8K dataset.
- Presented the paper at ICRCSIT-2020 online conference.
|
- Implemented a Python-based blockchain API with encryption.
- Built a complete web application with HTML, CSS and JavaScript on the front-end and PHP on the backend
- Tools used: Python3, Flask, Heroku, Cryptography python module, Firebase, HTML, CSS, Bootstrap, JavaScript, JQuery, AJAX, PHP, 000webhost.
|
- Created a custom web scraper to collect data on a few products periodically. Tools used: Python, Scrapy, PostgreSQL, Heroku.
- Performed data preprocessing - cleaning, normalisation, standardisation and generated several visualisations.
- Performed a few hypothesis tests and checked for correlation using different plots and linear regression.
|
- A web solution for airlines and travellers to conveniently track luggages across different airlines and countries.
- Since the data is stored on a decentralised blockchain, it will free the airlines from the burden of co-ordinating with other airlines and countries.
- Used Solidity, React, Web3.js, Truffle, Web3.py, Flask and some frontend design.
|
- Classification and Detection of Abnormalities in lung and bone X-Ray images.
- The model also accepted numerical and categorical inputs such as patient age and gender.
|
- A basic Web Application made by integrating a web server (Apache), an application server (Django), a database server (MySQL), a loadbalancer (HAProxy) and a firewall (ufw).
- Different servers were configured and integrated to create a multi-server architecture.
|
- A Heroku web app which extracts text from images of printed recipes or screenshots of a recipe.
- NLP is performed on the text and images, videos, GIFs are displayed based on the ingredients and steps.
|
- A (pure) C program that gets random tweets from the Twitter API and performs basic sentiment analysis on it using a dataset of words and associated sentiment.
- Tools used: C programming language, libcurl, libssl, Twitter API, SentiWords dataset.
|
- A python program that extracts the browser history of the user. It then categorizes the websites and displays statistics using graphs.
- Finally it tries to predict the gender of the user (prototype - uses a very small dataset).
- Tools used: python, pandas, matplotlib, sklearn
|
