Hashir Fida

I’m a Machine Learning Engineer designing agentic Large Language Model (LLM) workflows for document-wide legal edits driven by user instructions. At Nymble, I improved orchestrator accuracy by 15% using an embedding-retrieval-based toolset and increased revision quality to 87% semantic equivalence with professional lawyer edits through prompt engineering in an LLM-as-a-Judge framework.

I also build end-to-end ML products, like a React Native migraine prediction app using an NHS dataset, where I selected 28 high-correlation features and achieved a 77% F1 score, deploying the model on a Flask backend and integrating MongoDB for real-time inference. Previously, I developed a BERT-based Twitter hate speech classifier, improving performance by fine-tuning transformers and reducing class bias via undersampling to raise minority class representation from 12% to 50%.