Nutrition Management

Oryn-Nutri

VLM for intuitive, precise, and sustainable nutrition management.

Overview

Oryn-Nutri is a visual nutrition analysis model built for Speediance health and nutrition management scenarios. The model takes meal images or descriptions as input and outputs energy (kcal) and protein, fat, and carbohydrates (g) for each dish and for the meal as a whole. In the product stack, Oryn-Nutri serves as the nutrition perception component for food logging, providing a computable basis for recovery adequacy and macronutrient structure.

Core capabilities

Visual Recognition and Ingredient Decomposition

On the vision side, the model recognizes multiple dishes and, based on color, morphology, texture, cutting style, cooking traces, and common dish compositions, grounds nutrition estimation in an ingredient-level representation rather than a single dish label.

Nutrition Density Retrieval and Validation

The model interfaces with the USDA nutrition database to retrieve candidate nutrition-density entries, then validates and selects among them along three axes: food identity match, raw/cooked consistency, and processing-state compatibility. When candidates are clearly mismatched, a fallback mechanism is used to avoid accepting entries by default, so that the density entry and the weighing convention remain aligned under the same state.

Tool-Based Nutrition Calculation

Nutrition multiplications and summations are performed by a dedicated calculation tool: the model supplies each dish’s ingredient names, estimated weights, and selected densities; the tool executes the multiplications and summations and returns energy and the three macronutrients. Separating visual understanding from numerical computation improves the stability and auditability of long pipelines.