viraj sawad

I’m a Computer Engineering student working at the intersection of LLMs, multi-agent systems, and vision-language models, with a strong curiosity about whether models can actually reason. I’m currently exploring graph-based retrieval and self-supervised learning, aiming to turn ideas into systems that improve performance and reliability in real settings.

As an AI Fellow in the Fusemachines 2026 Nepal cohort, I’m expanding my depth across ML, Deep Learning, GenAI, NLP, Computer Vision, and MLOps through a capstone project. I’ve built projects like NepLegalKG for hierarchical GraphRAG over Nepal’s legal corpus, DR-Screen with a SimCLR-based self-supervised pipeline plus Monte Carlo Dropout for uncertainty quantification, and Arc—an education-focused multi-agent platform orchestrated via LangGraph shared-state graphs. I’ve also published a lightweight parasitic egg detection model using a modified YOLOv11 with Ghost Convolutions and hybrid SimAM-ECA attention, cutting computational cost by over 50% while boosting accuracy.