At NexSewa, I built and deployed NLP and computer vision models to production using PyTorch, FastAPI, and Docker. The internal API serves 3+ teams, and my automated data pipelines reduced manual data-prep time by roughly 40%.
I trained, evaluated, and optimized models at NexSewa, improving baseline accuracy by 8–12% through hyperparameter tuning and transfer learning. During the AI Fellowship, I built end-to-end projects in text classification, image classification, sentiment analysis, and object detection.
On my Nyaya Sathi project, I developed a legal AI assistant with a conditional RAG pipeline and a fine-tuned Mistral-7B model. I also built a soccer match analysis project that used computer vision and tracking to calculate player-level metrics.

