Projects
Research Paper - Rewiring Networks for Graph Neural Network Training Using Discrete
Geometry
- Implemented a library in Python using PyTorch Geometric library for Graph Neural Network (GNN)
training.
- Used deep concepts in differential geometry to prevent over-squashing of information between
long-range dependencies in
graph data.
- Achieved state-of-the-art results for several benchmarking datasets.
- Published by Springer at the Complex Networks 2023 conference.
- Presented the results of the paper on the international conference Information Geometry for
Data Science (IG4DS).
Technologies used: Python, PyTorch, PyTorch Geometric, NetworkX, Git
Paper link (Springer)
Paper
link (arXiv)
GitHub
link
MenuRadar Founder
- Developing a Google Maps-like application for searching for specific dishes across restaurants.
- Using GPT-4o (vision) to parse menu PDFs into JSONs and upload them to DynamoDB for user search.
Technologies used: Python, Langchain, DynamoDB, AWS
Stochastic Segmentation Library
- For an academic project, in a team of 6, implemented a Python Stochastic Segmentation Library
for detecting brain tumors.
- Under the supervision of Prof. Bernhard Kainz and two PhD students, implemented neural
network architectures like
Probabilistic U-Net and Probabilistic Hierarchical Segmentation (PHiSeg) based on Deep
Learning research papers.
- Recognised as one of the best 6 projects out of roughly 50 Computing group projects.
Technologies used: Python, PyTorch, Torchvision, MyPy, Tensorboard, Git
GitHub
link
IC Hack 20 - Poser
-
In a team of 4, built an application using Android Studio and Machine Learning models
for
Pose Estimation that helps a user exercise properly.
-
The app monitors all relevant angles centered at person's joints (with a phone camera) for a
selected exercise.
-
For each of the 3 implemented exercises, the app counts repetitions and gives voice warning
when a repetition is done incorrectly, e.g. "keep your arms straight", "don't move your
hips" etc.
-
Worked on implementing the angles for the exercises and counting repetitions.
Technologies used: Java, Android Studio, Machine Learning models
Devpost
link
GitHub link
IC Hack 19 winner - ARnold
-
In a team of 5, created an Android application using Unity Engine for visualising 3D models
in Augmented Reality (AR) during lectures.
-
The app allowed a lecturer to send a 3D model on a server that students could access with
their smartphones and view the model in AR.
-
The lecturer was able to highlight parts of the model and show labels with names of the
parts, as well as rotate the model.
-
Worked on interactions with the model, i.e. highlighting parts of the model and showing name
labels.
- Won the IC Hack 19 hackathon organised by Imperial College (Best Native Mobile App
category).
Technologies used: C#, Unity, AR, Git
Devpost
link
GitHub
link