DocSeeker: Structured Visual Reasoning with Evidence Grounding for Long Document Understanding | ScienceToStartup