Hello Everyone, these are few of the many projects i have built and developed.
Realtime Person Tracking and Re-Identification in Embodied Learning environment (Ongoing)
Created and annotated a person re-identification dataset for closed-room environment in a embodied learning class. Enhanced accuracy by fine-tuning pre-trained models (TriNet, Siamese networks, OSNet) specific to closed settings. Integrated Re-ID models with advanced object detection (YOLOv8) and motion tracking (Kalman Filters, Particle Filter, DeepSORT)
Privacy Protection of Student Video Data through Stylegan-Based Deidentification in Diverse Learning Environments (Ongoing)
Using various computer vision methods to explore innovative approaches to safeguard student privacy and confidentiality within various educational settings.
Protein Fold Recognition
Implemented advanced NLP techniques to improve protein fold recognition in challenging datasets (DD, EDD, TG, SCOPe) with diverse amino acid sequences and folds. Extracted features using evolutionary PSSM, HMM profiles, and concatenated them with global Convolutional and Skip Bi-gram features. Applied BERT and ESM by Meta for classification, achieving over 93% accuracy across all datasets, surpassing the previous 85% benchmark. Code and the paper will be released soon.
PEcAn SDA and Forecasting Dashboard
This project is primarily focused on creating an interactive display of the carbon cycle forecast and data assimilation system of PEcAn. The main goal is to revive the site-specific R Shiny Forecasting and SDA dashboard that is no longer operational and broaden its scope to cover a larger number of sites. Additionally, the plan is to construct a new comprehensive dashboard that combines both SDA and Forecasting graphs. Moreover, the proposal also aims to establish an automated notification email system for the visualizations.
More information can be found here
Extending APIs/ Distributed File Sharing for PEcAn Project
The project aims at improving and extending the current PEcAn REST API’s by making it more robust and dynamic from the user’s perspective. The security of the API’s will also be improved by introducing API KEY management system, rate limitng and input validation.
More information can be found here
Affect Aware Tutoring System in E-Learning systems (Pekanu E-Tutor)
Worked on a novel idea in the field of education technology, named ”Affect Aware Tutoring System Using Video Bots”.Built a learning management system that collects the click-stream log data of the student, simultaneously captures video and then predicts the user’s affect state for realtime feedback. Developed an optimized transformer-based deep learning model Vision Transformers to predict the user’s affect state. The model was trained on the huge DAiSEE dataset for approximately 300 hours. Paper documented and to be submitted in IEEE Transactions in Learning Technologies. The code will be open sourced soon.
Covid-19 Mortality Prediction
Data analysis and processing: Proficient in cleaning, exploring, and transforming private hospital data with 1 lakh data points to better understand and extract information using Pandas and Python. Data Imbalance Mitigation: Employed techniques such as SMOTE and undersampling to effectively tackle the challenge of imbalanced data. Mortality Prediction: Implemented state-of-the-art machine learning models to predict COVID-19 patient mortality, utilizing an ensemble approach and achieving an impressive accuracy of 93%
Implementation of BERT language model in Rasa NLU
Implementation of BERT language model in Rasa NLU to build a general purpose contextual chatbot with good precision. It is a part of the Humanoid project at IIIT Dharwad. Report can be found here.
A Meta-Analysis of Probiotic Efficacy for Gastrointestinal Diseases.
Meta-analyses on the effects of probiotics on specific gastrointestinal diseases have generally shown positive effects on disease prevention and treatment. Probiotics are products or preparations that contain enough living bacteria to affect the microflora communities of a host. Report can be found here.
Optimised Blockchain Based Fair Payments System to Outsource Cloud Computation Tasks.
A computational website to perform various deterministic computation without a trusted third party. Performed Case study, proposed improvements and implemented the paper. The code can be found here.
Karnataka Telemedicine Utilization
A site made in php that runs a python script to visualize user uploaded dataset of Karnataka healthcare facilities in the Karnataka state map using EDA techniques written in python.
Geo-Health Karnataka
Funded project under Dept. of Health and Family Welfare Services, Govt. of Karnataka. Geo Health is an application to display all of the Karnataka state’s healthcare facilities on map to help citizens get free and fast access to healthcare information.
Camp+
The application Camp+ which is built in Flutter and NodeJS manages deliveries, visitors to the campus, helps the residents of the hostel in viewing and calling other residents of different hostels, allows booking housekeeping facilities and solves several other essential needs relevant to residential campuses like library booking, notice board management, cafeteria slot management, event booking management, room booking management, laundromat slot booking, canteen delivery system and so on. I worked in the backend which included working with NodeJS in compliance with the MVC module structure for developing various modules for the features. AWS services like S3 and DynamoDB and also MongoDB were used for data storage. For more information please check out the Camp+ website.
Chitran
The “Chitran” platform allows the user to view the various crime data segregated in a very structured manner. As a part of this work, I facilitated the automation of cleaning huge chunks of data provided in the National Crime Records Bureau (NCRB), GoI website. Furthermore, after the cleaning I had to label encode the data for converting them into the machine readable form. Eventually after the preprocessing of the data it was time to visualize the data into various charts which are very easy to understand.
Sarathi (Safe Zone Detection for UAV)
I worked on the safety and backup mechanisms of the Unmanned Aerial Vehicles (UAV) in case of crash landings, loss of connection, system failure, loss of control etc. The system uses compression and sensing techniques to detect a safe zone using HD cameras fitted onto the UAVs. Compressed Sensing will benefit us in saving Power, low size of the processed image and efficient use of Bandwidth for communication. I worked on the Gabor filters, Compressed Sensing, Markov Chain Code and Chi Square Techniques.
Terra&Tech (ML-based house predictor)
It is a Machine Learning based house predictor for the people who buy houses taking a Home Loan.
Program is built in python environment using python libraries and machine learning concepts. The dataset is cleeaned repeatedely so that it can be readily used by the python program to apply machine learning algorithms. After going through number of regression techniques , finally multiple linear regression algorithm fit close to the dataset. once we have finished the python program we designed API files for the same and took it to the web interface to be readily used in a new or existing systems. The program takes input of your budget and other financial conditions and then gives you the output of the most suitable assets you may own with other benefits. You can test out the website here: Terra&Tech
Fruits Image Classifier using VGG16
I used the Fruits 360 dataset and VGG16 which is a pretrained CNN. Data Augmentation techniques were also implemented in order to fit the model more accurately. The fitting takes a lot of computational power and time.
THe Kaggle notebook for the code can be found here: Fruits Image Classifier using VGG16
Covid Tracker
Tracks the information regarding the Covid-19 essential for a student to know before he or she finds a job in the affected places.
For more information about the project and the code you can check out the Github