Ryan Witzman
I build intelligent systems end-to-end — RAG pipelines, fine-tuned LLMs & rerankers, and computer vision models deployed on real workloads. Currently engineering biomedical data pipelines at the Indiana Center for Musculoskeletal Health.
01 · Experience & Education
Building an automated pipeline for musculoskeletal research data and running neural-network inference over large biomedical video datasets to extract structured signals for downstream ML.
Fine-tuned rerankers inside a production RAG stack for an AI study platform used by hundreds of thousands of students — lifted document retrieval performance +20%. Built backend APIs wiring the full stack together.
Bachelor of Science in Artificial Intelligence with a minor in Mathematics. Focused on ML architectures, optimization, and applied research.
02 · Selected Work
Aerial-drone object detection model — 0.93 mAP@50, 14ms inference. Custom dataset curation + lightweight architecture for edge deployment.
17M-parameter decoder-only transformer trained from scratch. Learns 5-7-5 structure and semantic coherence on a curated haiku corpus.
YOLOv11s tuned on A100s for vintage pinball displays under glare and reflections. 0.99 mAP@50 across 1.1k labeled frames.