×

Trans Lingua Script

Developed a model which can detect the signboards, extract the text from the signboards and convert it into Hindi language.

Developed an object detection model to detect sign board objects in a given image and a CNN-RNN encoder decoder model to convert the cropped signboard into text and later used a sequence to sequence with attention model for transliterating the above sequence of characters from English to Hindi language.

Achieved an accuracy of around 90% and developed using Python, TensorFlow, Keras and Natural Language Processing.

Project Link

×

"EmoSynth: Expressive Avatar Generation with Deep Learning"

Developed a model which is capable of creating emojis by analyzing the live feeds in real-time to capture the face, to find out the expression, to classify the expression and map to corresponding avatar.

Used a convolution neural network architecture and trained on the FER2013 dataset for emotion recognition from images and achieved an accuracy of around 86% by using Python, TensorFlow, Keras and OpenCV.

Project Link

×

Text2Emote

Deep Learning for Emoji Prediction

Developed a neural network model to identify the relationship between text messages and the emojis used in them and to use natural language processing techniques to analyze the text sentence's underlying semantics to predict appropriate emojis based on the context.

Implemented Bidirectional LSTM, LSTM, Support Vector Classifier, and Naïve Bayes models out of which Bidirectional LSTMs performed better with an accuracy of 89%.

Used tools and technology such as Python, TensorFlow, Keras, and Natural Language Processing.

Project Link

×

Resell Radar

Developed a predictive model for pre-owned car prices by implementing Random Forest, Linear Regression, Ridge Regression, Lasso, KNN, and XGBoost models, resulting in an attained accuracy level of approximately 85%.

Executed rigorous data pre-processing, including feature engineering and selection for improved pattern capture.

Selected models systematically based on MAE, RMSE, and R2 score, optimizing accuracy through hyper-parameter tuning.

Recognized among the top real-time projects for Applied Numerical Methods course in 7th semester.

Tools Used: Python, Pandas, Numpy, SciKit-learn, Flask, SQL, HTML, CSS, Git.

Project Link

×

TweetTone

Implemented and compared various ML algorithms to determine the tone of the text whether it is positive or negative.

Used approximately 1.6 million tweets extracted from the twitter API as the dataset.

Implemented few of the most popular algorithms like Logistic regression, Multi layer perceptron, Linear support vector classifier.

Used Accuracy, F1score, cv_scores as performance measures.

Project Link

×

Microstructural Characterization and Mechanical Performance of AZ91 Magnesium Alloy Processed by Friction Stir Processing Using Novel Tool Designs

Worked in a team of 2 and developed a novel tool to study the pulsating effect of number of faces on the mechanical and metallurgical behavior of friction stir processed A91 alloy.

Observed the improvements in properties in materials processed with the newly proposed tool.

Carried research under the guidance of Dr. Jose Immanuel, Assistant Professor, Department of Mechanical Engineering.

Publication

×

Influence of titanium particulate reinforcement on microstructural evolution and mechanical performance of AZ91 magnesium matrix surface composite developed through friction stir processing

Developed a model which is capable of creating emojis by analyzing the live feeds in real-time to capture the face, to find out the expression, to classify the expression and map to corresponding avatar.

Used a convolution neural network architecture and trained on the FER2013 dataset for emotion recognition from images and achieved an accuracy of around 86% by using Python, TensorFlow, Keras and OpenCV.

Publication

×

Development of correlation between temperature, liquid life span, molten pool, and porosity during Wire Arc Additive Manufacturing: A finite element approach

Developed a neural network model to identify the relationship between text messages and the emojis used in them and to use natural language processing techniques to analyze the text sentence's underlying semantics to predict appropriate emojis based on the context.

Implemented Bidirectional LSTM, LSTM, Support Vector Classifier, and Naïve Bayes models out of which Bidirectional LSTMs performed better with an accuracy of 89%.

Used tools and technology such as Python, TensorFlow, Keras, and Natural Language Processing.

Publication

×

Design and simulation of roll cage of an all-terrain vehicle

Developed a neural network model to identify the relationship between text messages and the emojis used in them and to use natural language processing techniques to analyze the text sentence's underlying semantics to predict appropriate emojis based on the context.

Implemented Bidirectional LSTM, LSTM, Support Vector Classifier, and Naïve Bayes models out of which Bidirectional LSTMs performed better with an accuracy of 89%.

Used tools and technology such as Python, TensorFlow, Keras, and Natural Language Processing.

Publication

×

Use of Hydrogen in IC Engines Study on Performance, Emission and Fuel Injection System

Developed a neural network model to identify the relationship between text messages and the emojis used in them and to use natural language processing techniques to analyze the text sentence's underlying semantics to predict appropriate emojis based on the context.

Implemented Bidirectional LSTM, LSTM, Support Vector Classifier, and Naïve Bayes models out of which Bidirectional LSTMs performed better with an accuracy of 89%.

Used tools and technology such as Python, TensorFlow, Keras, and Natural Language Processing.

Publication