Editorial nutrition · Open catalog
An AI import pipeline turns YouTube transcripts, recipe URLs, and typed notes into USDA-resolved ingredients. Iterative algorithms score meal plans against your macro and micronutrient targets — down to the gram.
01 · Import pipeline
Drop a YouTube URL, a recipe page, or type ingredients by hand. The pipeline transcribes, extracts, resolves each ingredient against a local catalog, then USDA, then an AI fallback — and aggregates macros and 30+ micros.
Transcript-driven ingredient extraction. Handles chef-style narration, unit conversion, and implicit quantities.
Any recipe site. Structured data first, HTML fallback second, always sanitized before storage.
Ingredients fall through: local catalog → USDA FoodData → AI estimator. Every gram is accounted for.
02 · Meal-plan algorithms
Phase 3 optimizes one meal-block at a time against a daily nutrient target — never averaging away a shortfall. You get three graded variants to compare and flip through.
Balanced
Dinner → lunch → breakfast. Fastest plan, hits macros hard, minimal drift on micros.
Adventurous
Same greedy core with cuisine-repetition penalties. Broader flavor spread across the week.
Optimized
Stochastic search across candidate combinations. Best nutrient match, slowest to converge.
Zero-centered delta bars across every macro and micro. Grade is per-variant, not averaged.
Shortfalls become an AI prompt for raw-ingredient sides. Each side is USDA-resolved and re-graded.
03 · Your kitchen
Everything personal is end-to-end encrypted with a key that lives on your device. The catalog stays public — your pantry never does.
Side and snack ingredients expand into per-day gram totals. The list matches what the algorithm actually planned.
Direct meal links, quick removal, and a full wipe when you want to start clean.
The catalog