P/01
Projects
Reproducible research code, data analysis, simulation, and machine-learning experiments—documented through public GitHub repositories and Kaggle notebooks.
dipul@lab:~/projects$ find . -maxdepth 1 -type project
status: 4 public records indexed
Public repository index
Each entry links to its source and states what the repository currently contains—without treating planned work as a finished result.
A duty-cycle-correct simulation of post-quantum ML-KEM handshake completion over LoRaWAN under real EU868 enforcement.
Repository scope: four deterministic configurations under an ideal channel. It models one direction of the fragmented exchange and does not yet include packet loss or multi-device contention.
A reproducible R pipeline comparing differential gene expression, enrichment, survival, and protein-interaction networks across TCGA-BRCA and TCGA-LUAD.
The discovery analysis uses 200 randomly sampled tumors per cancer to keep the workflow tractable on personal hardware; the repository documents this as a limitation.
An end-to-end transfer-learning project for classifying dermatoscopic images into seven HAM10000 lesion categories through a Streamlit interface.
Educational research only—not a certified medical device and not suitable for diagnosis. The repository does not publish a trained model or completed accuracy figures.
A notebook-level pilot for the Green AI Reclamation Framework, comparing reclaim-first and scale-first approaches across GPU energy, wall-clock time, validation loss, and peak-memory constraints.
The research visual records a GPT-2 medium / WikiText-2 experiment on a single NVIDIA T4 GPU; broader hardware and workload validation remains future work.
Source profiles
Collaborate