Ramon User

At Unstructured, I build Python-based document partitioning, metadata extraction, enrichment, embedding, and retrieval pipelines that turn heterogeneous enterprise documents into reliable structured data and retrieval context for AI workloads.

My doctoral work at Northeastern University focused on machine-learning systems, GPU-accelerated inference, and hardware-aware model optimization. I built CUDA/C++ kernels and PyTorch extensions for operator fusion, mixed precision, tensor transformations, and GPU-memory optimization, with reproducible evaluation across quality, latency, throughput, stability, and resource utilization.

Across more than 25 years in software, data, and platform engineering—including MRI Software, LeanDog, and Hyland Software—I've built data services, integrations, scalable processing infrastructure, and customer-facing applications. I value research rigor alongside production engineering: AI should be measurable, reproducible, deployable, monitored, trusted, and useful.