I published AffectFusion on Zenodo, comparing early and late fusion architectures for multimodal emotion recognition. My study evaluated audio-visual models across cross-dataset and synchronized within-dataset settings.
For AffectFusion, I developed Gated Feature Fusion, Multimodal Self-Attention Fusion, and a Smart Hybrid Fusion mechanism. In the cross-dataset evaluation, early fusion reached 66.88% accuracy, compared with 64.80% for late fusion.
I also developed a custom U-Net for multimodal brain tumor segmentation on BraTS2021 and built an end-to-end MRI preprocessing and evaluation pipeline. I deployed a standalone segmentation model and an interactive BraTS2021 demo on Hugging Face Spaces.
At Capgemini, I built and maintained Python REST APIs and backend systems as a Senior Software Engineer, improving average API response time from 450ms to 180ms. Earlier, as a Software Engineer at Capgemini, I developed Java/MongoDB microservices and supported CI/CD pipelines on Azure DevOps.
