Hi There! 👋 I am
Kusha
I'm a dedicated Backend Developer based out of Gurugram/Delhi specializing in algorithms and data structures with > 7 years of experience, where I thrive on turning complex software challenges into scalable solutions.
Connect with me! Resume Checkout my reading list here! Medium Articles! Github
Projects
Developed an exoplanet detection system using ML on Kepler light curve data, extracting key time-series features from stellar brightness data to detect transits. Achieved promising results with two approaches: CNN-LSTM for direct time-series modeling, and feature engineering with MLP and XGBoost, leveraging custom spectral and temporal features, including transit dip distances. This project demonstrates AI’s potential to accelerate exoplanet discovery in large astronomical datasets.
Space debris poses a significant threat to satellites, space missions, and other orbital assets. This project aims to tackle two critical challenges in space debris management:
Debris Detection: Identifying space debris using machine learning models based on orbital parameters extracted from TLE (Two-Line Element) data.
Trajectory Prediction: Developing a system to predict debris trajectories using time-series analysis of orbital data.
By addressing these challenges, this project contributes to safer and more efficient management of Earth's orbital space.
Classification of activities in videos using machine learning using Motion History Images (MHI). MHIs were implemented using the frame differencing method, these were then converted to features using Hu moments.
Predicting CO2 emissions with onboard diagnostics (OBD) time series data from vehicles provided by the Envirocar Project and explores efficacy of variations of LSTM and GRU models in predicting CO2 emissions.
Simultaneous Localisation and Mapping
Programming a robot (left red dot) to navigate through an unknown territory of trees, avoiding crashing into them to reach the target location (red triangle on the right side)
Localisation of a satellite orbiting within a solar system using particle filters. White dots are where the system is trying to find it and the blueish dot is the estimated location. Particle filters have an advantage of easy implementation with approximate solutions for multi-modal problem statements.
Imagine you discover a new galaxy and you observe the colors of the galaxy (the relevant data), how would one know if it is redshifted or simply has a different spectrum? In this particular project, I used (Sklearn lib) Decision Trees with k-cross-validation to train and predict Redshift.
The project involved creating a Knowledge based AI system to solve RPM assessments. Its starts image processing to create features in a figure, subsequently creating relationship features between the set of figures and then using them to predict the answer.
Just a bunch of subrepos which target kaggle starter projects (digit recogniser, titanic etc)
