At Nysonian, I fine-tune and deploy a Gemma-based sentiment classification model on AWS ECS for real-time customer ticket prioritization. I also maintain data pipelines integrating Shopify, ShipHero, and MongoDB across four storefronts.
At CareCloud, I fine-tuned Llama 2 7B with QLoRA/PEFT on clinical data, improving entity extraction accuracy by 35% and reducing inference cost by 40% versus the GPT-4 API. I also built a MongoDB tagging pipeline that processed approximately 377,000 closed support conversations.
My research includes a self-supervised Vision Transformer Autoencoder for X-ray threat segmentation and a CodeBERT study benchmarking seven transformer architectures on bug reports. I have five peer-reviewed publications spanning computer vision, NLP, and signal processing.
