»PROJECTS
Geothermal Heat Pump Simulation
Description: Developed a simulator for geothermal heat pump cycles to analyze component and setup dependencies. Simulates fluid flow and heat conduction through generic pipe geometries and transport media.
- Implemented pipe environment and visualization pipeline in Python
- Derived and solved temperature propagation equations along complex paths
- Implemented numerical integration for heat propagation in C
ODE-Solver Software
Description: Developed a flexible and extensible C++ solver for ordinary differential equations (ODEs), supporting implicit and explicit methods including Runge-Kutta, Adams-Bashforth, and Euler. (Toy example: RK4 for planetary motion — image from here)
- Implemented solver using object-oriented programming in C++
- Designed architecture for easy extension and method customization
- Configured build system with CMake and Makefiles
- Developed test suite using Google Test (gtest)
Aerial Road Segmentation
Description: In this project, we evaluate the performance of various convolutional neural network (CNN) based architectures for road segmentation from aerial imagery.
- Used Hydra and PyTorch
- Implemented U-Net and ResNet architectures
- Fine-tuned pretrained segmentation models
- Experimented with image transformations and patch sizes
Neuronal cell morphology classification
Description: Combined Algebraic Topology and Graph Neural Networks (GNNs) in multi-embedding fusion models for classifying neuronal morphologies.
- Implemented multi-embedding fusion model using PyTorch, and PyTorch Geometric
- Generated topological representations via Persistent Homology
- Developed pipeline to reproduce state-of-the-art research in neuronal morphology classification
- Improved classification performance by combining GNN and topological features
Graph based Epilepsy Seizure Detection form EEG-Signal
Description: Explored Graph Machine Learning for seizure detection, experimenting with signal representations, graph construction, and model architectures.
- Built feature processing pipeline in Python
- Implemented benchmark models for multi-channel time series, including CNNs and LSTMs
- Improved classification performance using Graph Neural Networks and hybrid approaches
Feature Engineering on BRFSS survey data for Heart Attack prediction
Description: Predicted heart attacks from survey data using feature engineering and logistic regression models.
- Built data cleaning, representation, and augmentation pipeline with Python and pandas
- Developed prediction pipeline with Python and PyTorch
