Projects
Here are selected projects that best represent my technical depth (modeling + implementation + evaluation).
Open-source
GNN-Keras3 (TensorFlow/Keras)
Graph Neural Networks library with a Keras-style API for training/inference.
- Layers: graph convolution, graph attention, differentiable pooling, etc.
- Includes runnable examples (e.g., protein graph classification; ~80% single-fold validation accuracy on TUProteins in one demo).
- Repo: https://github.com/arijitcodespace/GNN-Keras3
- Docs: Get Started
Research / academic projects
Convergence of Asynchronous SGD for PL functions
Theory + experiments around asynchronous SGD under the Polyak–Łojasiewicz condition.
- Paper PDF: link
- Code: https://github.com/arijitcodespace/Asynchronous-SGD
Text-to-Image Synthesis Optimization (UCLA MS project)
Optimized text-to-image GAN architectures under hardware constraints; integrated CLIP embeddings for stronger text conditioning.
- Achieved FID 9.85 with a 10.6M parameter generator (as a lightweight alternative to larger baselines).
Mini-GPT (TinyStories)
Implemented a small GPT for story generation (course project), including parameter-saving tricks like weight tying.
- Top-3 token accuracy: ~88.6% on TinyStories.
VOC signature clustering (B.E. thesis)
Machine-learning approach for clustering signatures of volatile organic compounds (VOCs) using E-Nose signals as a low-cost alternative to spectrometry.
- Achieved ~91% accuracy and ~0.89 F1 using a hierarchical pipeline (vs. weaker baselines).
Want a shorter view? The Home page has a quick list; this page is the “details” version.