{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/c56ad021969f48bc95c4ec595afff54f\" frameborder=\"0\" width=\"2080\" height=\"1560\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1560,"width":2080,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1560,"thumbnail_width":2080,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/c56ad021969f48bc95c4ec595afff54f-1642a0ea2cd2946c.gif","duration":121.466667,"title":"Meal Measure Hackathon: Verified Nutrition Estimates","description":"This Loom explains how the meal measurement app estimates food portions and computes nutrition with auditable checks. It uses Sam 3.1 to segment a plate, while Mew spark names items and estimates a grams-based portion range without writing calorie numbers, then near-misses are one tap away and everything recomputes. The app includes an optional plate weight anchor but flags disagreements between scale and OCR, for example 545 grams vs 245, and requires checks before anything becomes a fact. For accuracy, the author audited seven real meals component by component and published errors, with media and total error around 31% and specific miss patterns like milk under cereal and mismatched strawberry pricing."}