Summary
Hunter Heidenreich is a Senior AI Research Scientist with a decade of experience building production-scale language and vision models, currently leading VLM research and distributed DGX H100 training at Roots.ai. He has a strong academic foundation from Harvard and Drexel, where he developed generative surrogate models for molecular dynamics and fast-FFT compiler work, framing a recurring research question: how to represent data that straddles continuous and discrete domains. At Roots he led GutenOCR and the 1.5M-page PubMed-OCR release, beating prior open-source OCR baselines and operationalizing terabyte-scale multimodal training and high-throughput long-document inference optimizations. His work blends principled research (publications in COLING, AIES and arXiv) with hands-on engineering: driving extraction accuracy from <50% to >90% in production and tuning vLLM stacks for 4–5x throughput gains. He maintains an open research profile with public notes on hundreds of papers and emphasizes dataset quality, cross-validation, and invariant enforcement as levers for model reliability. An uncommon throughline in his career is treating symbolic encodings like SMILES as potential distinct modalities and designing tokenizers and architectures around that insight.
9 years of coding experience
8 years of employment as a software developer
Bachelor of Science (B.S.), Computer Science, 4.00, Bachelor of Science (B.S.), Computer Science, 4.00 at Drexel University
Master of Science - MS, Computer Science, 3.94, Master of Science - MS, Computer Science, 3.94 at Harvard University
English