Omer Nasir

Developed an end-to-end speech analytics pipeline achieving 96% transcription accuracy, automated opportunity detection and intelligent alerting to reduce revenue loss, and delivered a React dashboard for live performance tracking and call quality assessment.

Architected an AI-powered psychiatric analysis platform that reduced manual transcription time from 8 hours to 20 minutes per session and automated clinical note generation to cut documentation time by 85%. Implemented local LLM inference, robust SSO (OIDC/SAML) over private VPN, and achieved 97%+ transcription/diarization accuracy with 99.5% uptime for 25+ healthcare staff.

Built self-supervised pretraining using SwinV2-Large with iBOT and implemented Mask2Former in Detectron2 for microscopy segmentation, enabling scalable distributed experiments and automated batch inference for cell area calculations.

Assisted Linear Algebra, Operating Systems, and programming labs, supporting 100+ students and mentoring on Python, C, and software development best practices across coursework and lab sessions.

Built production-ready RESTful APIs with FastAPI and deployed scalable applications on AWS EC2 with MongoDB, improving backend efficiency by 25% and delivering documented endpoints with robust error handling.

Developed an LLM-powered legal chatbot with RAG architecture using Pinecone and FastAPI, improving response accuracy by 40% and reducing API latency by 30% through optimization and low-latency deployment.

Built a transformer-based Urdu-to-English proverb translation model using MarianMT and created a 2,000+ pair custom dataset, improving translation accuracy by 70% for domain-specific tasks.

Bachelor of Science in Computer Science with coursework in Machine Learning, Deep Learning, Natural Language Processing, Data Structures, Algorithms, Linear Algebra, and Operating Systems.