Voice User Interface

September 2018 − October 2018

Implemented code to take raw audio as input and return a predicted transcription of the spoken language.
Created several Deep Neural Nets models with various architectures to develop an efficient network.
The final system consisted of two Conv Nets followed by a bidirectional RNN and a Time Distributed layer.

Machine Translation

September 2018 − October 2018

Developed a Deep Learning system to translate English to French.
Implemented several Deep Neural Networks with different configurations to find a suitable architecture.
The final custom model created consisted of multiple bidirectional RNNs and Time Distributed layers.

Part of Speech Tagging

September 2018 − October 2018

Developed a speech tagger to determine the syntactic category of the word from the words in the surrounding context.
Used the Pomegranate library to implemented Hidden Markov Models (HMM) for part of speech tagging on Universal Tagset.
Contrasted the tagger's performance with 'Most Frequent Class' Tagger.
The developed HMM achieved training accuracy of 97.54% and test accuracy of 95.96%.

Forward Planning Agent

August 2018 − September 2018

Implemented a search agent to solve planning problems and compared with other search algorithms.
Used symbolic logic and classical search to deploy an agent that performs progression search to solve scheduling problems.
Analyzed and tabulated the agent's performance with other search algorithms like Breadth First Search, Depth First Search, Uniform Cost Search and Greedy Best First Search on parameters like time, expansions and path length.

Sudoku Solver

August 2018 − September 2018

Landmark Detection and Robot Tracking

April 2018 − June 2018

Implemented "Simultaneous Localization and Mapping" (SLAM) algorithm for a two-dimensional world.
Implemented the code to read the sensor values providing surrounding information and govern the mechanics of the robot.
Implemented the SLAM algorithm to get accurate coordinates of the landmark.

Image Captioning

April 2018 − June 2018

Created a system that could automatically caption images using Deep Learning techniques.
Implemented Encoder − Decoder architecture consisting of CNNs and LSTM layers for Encoder and Decoder respectively.
The Encoder CNNs were used to learn the image features while the Decoder LSTMs provided the requisite captions.

Facial Key points Detection

April 2018 − June 2018

Developed a system to detect key facial features like eyes, nose, and mouth.
Used Computer Vision techniques and Deep Neural Networks to create a facial recognizer on par with the Haar Cascade.
Developed a Deep Neural Network consisting of multiple Convolutional layers, maxpool layers, and fully-connected layers.
The fully trained system could identify the aforementioned features using 64 unique points.

Object Classification

March 2018 − April 2018

Applied Deep Neural Networks (DNNs) for classifying objects.
Implemented a convolutional neural network to classify images from the CIFAR-10 dataset.
The model achieved an accuracy value of 72.3 %.

Smart Cab

March 2018 − April 2018

Applied reinforcement learning techniques for a self-driving agent in a simplified world to aid it effectively reach its destination in the allotted time.
This project involved modeling a complex control problem concerning limited available inputs and designing a scheme to automatically learn an optimal driving strategy based on rewards and penalties.

Customer Segments

February 2018 − March 2018

Implemented clustering algorithms on unstructured data to understand the patterns and natural categories in the data.
Used multiple clustering algorithms and compared their results empirically.
Made predictions about the logical categories of various types in a dataset, then checked these predictions against the result of the unsupervised analysis.

Finding Donors for CharityML

February 2018 − March 2018

Investigated factors that affect the likelihood of charity donations being made based on real census data.
Trained and tested several supervised machine learning models on pre-processed census data to predict the possibility of contributions.
Selected the best model based on accuracy, a modified F-scoring metric, and algorithm efficiency.

Data Science Capstone

July 2017 − August 2017

Designed a machine learning based regression system to predict the Repayment rate.
Created a model that could predict the repayment rate for students based on the values provided in the dataset using Microsoft Azure Machine Learning.
Tested and developed various models to find the model with the lowest RMSE value and used Boosted Decision Trees for learning as they gave lowest RMSE value of 7.18

Character Recognition

November 2016 − May 2017

Developed a system to read English printed and handwritten characters and converted them into editable digital characters.
Used OpenCV library for image acquisition and processing and k-nearest neighbor algorithm for classification and solved the problem of image normalization for segmentation and feature extraction.
Developed the dataset and test set for the system which was able to read and correctly identify characters in approximately 80% of test images.

Home Automation using Arduino

January 2016 − April 2016

Created a remotely controlled switch using Arduino controller and an android app to control the switch.
Used Arduino microcontroller to create a switch to control electrical appliances and developed an android app using Android Studio to control the switch connected through Bluetooth.
Designed the scheme to connect the application with the Arduino microcontroller and created a prototype application using MIT app-inventor and used Android Studio for the final application.
The system was able to control electrical appliances with ease and was less energy consuming than other systems available as it used low power Bluetooth module for communication.

Projects