Projects

SAR Target Classification (J-CNN + TDP)

Designed a dual-branch CNN for SAR image despeckling and classification. Achieved 94% accuracy on the MSTAR dataset with 17.7% filter pruning via task-driven criteria.

References: [1] [2]

Band Gap Prediction Using DFT + ML

Created a predictive model for material band gaps using data from the Materials Project API with VASP and Python.

Lankford Coefficient Prediction

Predicted Lankford coefficients (r-values) from ODF images using CNN. Enabled faster estimation of plastic anisotropy in metals.

Stress Hotspot Prediction

Used microstructural texture and geometry-derived features to predict stress hotspots in FCC materials via ML models.

Causal ML: App Engagement Analysis

Used Causal Trees and Random Forests to estimate the treatment effect of rewards on app participation across user demographics.

Concall Sentiment Analysis

An NLP-based tool for analyzing sentiments in conference call transcripts using VADER and Afinn. Supports PDF/Excel input, tokenizes speech content, and exports normalized sentiment metrics and key word extractions to CSV.