Hybrid Recommender System

• The algorithm used combination of twitter sentiment analysis and content based filter.
• Both scores were fused to calculated final score.
• Trained model was deployed on Google Cloud App Engine and Flask REST API was used to connect requests from android app.
• Designed basic functionality of login and signup using firebase and saved the user's metadata on cloud bucket.

Facial Keypoint Detection

A Computer Vision problem where given an image, the facial keypoints are needed to be located

• It used combination of Haar Cascade and Convolutional Neural Network (CNN)
• CNN was trained to map 68 keypoints across a gray scale image of face
• Keypoints were translated back to original RGB image to map on face
• A Node.js baed web application was developed to show functionality
• With proper dataset CNN can be modified to predict emotions from facial expressions as well

Simultaneous Localization and Mapping (SLAM)

Simultaneous Localization and Mapping (SLAM) for creating a 2D map based on robot sensor data.

• It combines robot sensor measurements and movement to create a map of an environment from only sensor and motion data gathered by a robot, over time.
• SLAM gives you a way to track the location of a robot in the world in real-time and identify the locations of landmarks such as buildings, trees, rocks, and other world features.

Automatic Image Captioning

A Computer Vision problem where given an image, the model should generate caption for that image.

• A Convolutional Neural Network (CNN) is used as encoder for generating features vector to be used by decoder RNN
• LSTM based RNN takes feature vector from CNN and caption dataset is used to train the RNN
• Final trained model can generate captions given an image
• A Node.js web app was developed to show functionality

Computer Graphics and Image Processing

A collection of core computer graphics algorithms and image processing techniques.

• Computer Graphics Algorithms: 2D Transformations, Z Buffer, Flood Fill, B-Spline Curve etc.
• Image Processing: Image Gradient, Equalization, Masking & Filter etc.

DCGAN for Face Image Generation

A Deep Convolutional Generative Adversarial Network (DCGAN) implementation for face images generation.

• Model was trained on a modified version of CelebFaces Attributes Dataset (CelebA) having 32x32 size images.
• Implemented model was able to generate 32x32 images as output.
• Implemented model can be scaled up to train on higher resoultion images to generate higher resoultion images.

Reinforcement Learning Agents

RL Agents implemented to perform specified tasks.

• Unity ML Agents environment is used simulate the agents
• 1 - Double jointed robotic Arm for reaching object
• 2 - Tennis game between two players

Basics of CUDA C++

Implemented programs with CUDA which cover vector operation and other operations which can use parallel compute to speedup.

GA (Genetic Algorithm) for optimization problem (TSP & Advt.)

C++ implementation of GA for TSP problem. Evolutionary algorithms are used to get near optimum results by using multiple random search spaces and evolving the ones with best fitness. The algorithm implemented reached some better results for large number of vertices.

Socket Programming

A set of socket programs implemented in Java that cover RMI, Peer to Peer and Multi-threaded servers.

Lex-Yacc implementation to generate 3 address code

A set of programs that generate 3 address code given the grammar and syntax directed translation rules.

VLSI VHDL Programs to implement basic logical blocks

A set of programs that cover structural & behavioural model for basic logic blocks like Adder, Ripple Carry Adder, Comparator, Multiplier and Counter.