Data Sources for Nutritional Values

Summarize with: ChatGPT Claude AI Mode Perplexity

At NutriScan, providing accurate and reliable nutritional information is paramount. We utilize a multi-faceted approach to ensure the data you receive is as precise as possible, powering key features throughout the app.

Our methodology combines several high-quality sources:

1. Our own food database. Most lookups now resolve against a nutrition database we build and maintain ourselves, covering millions of foods. It is assembled from established public and licensed references and then cleaned in-house:
   • USDA FoodData Central for generic and whole foods.
   • Open Food Facts for packaged and barcoded products.
   • FatSecret for additional branded and regional coverage.
   • In-house review and correction. Imported records are not trusted as-is. We reconcile calories against protein, carbohydrate and fat using Atwater factors and correct rows where they disagree, resolve cooked-versus-raw and dry-versus-prepared basis errors, remove duplicate and malformed entries, and set realistic household portions (a slice, a cup, a piece) per food rather than applying one generic weight to everything. Food names are maintained in multiple languages so regional dishes resolve in the user's own language.
   • Building this ourselves is what lets us cover home-cooked and regional dishes, which barcode-first databases largely miss.
2. Advanced Internet Search: For unique, regional, or newly available food products not yet in our database, our systems perform targeted, high-quality searches across reputable online sources to gather the necessary nutritional information.
3. OpenAI API Integration: We leverage the power of the OpenAI API for several critical tasks:
   • Analyzing Scanned Meals: Helping to identify complex food items within an image and understand the context of the meal.
   • Cross-Referencing Data: Validating information found across different sources for improved accuracy.
   • Powering Diet Plans: Assisting in processing user preferences and health goals to generate personalized meal recommendations and nutritional targets.
   • Enhancing NutriBites: Understanding user queries about their meal history to provide relevant insights.

This robust combination allows us to provide reliable nutritional breakdowns across various app features:

• Scanned Meals (Process Meal Scan & Nutrition Details Page): When you scan a meal, image analysis identifies food items. Our system then queries the integrated data sources (our own food database first, then advanced search and OpenAI analysis) to retrieve macronutrients (Calories, Protein, Carbs, Fat), top micronutrients (like Fiber, Potassium, Calcium), and calculate the NutriScore. The process includes checks to handle non-food images or potential content filtering issues.
• Manually Logged Foods & Edits (Nutrition Details Page): When you manually log food, add missing items to a scanned meal, or adjust portion sizes, the app queries the same data sources to find and display the relevant nutritional information.
• Personalized Diet Plans (Diet plan - Premium Feature): Generating a diet plan involves combining your detailed profile (preferences, goals, restrictions, etc.) with our nutritional data retrieval methods. An LLM, primarily powered by OpenAI, processes this information to create daily macro/micronutrient targets and suggest specific meals with detailed breakdowns (recipes, portion sizes, benefits).
• Insights & NutriBites (Insights, NutriBites): Features like the monthly calendar view (showing NutriScore colors), daily nutritional breakdowns, and answers provided by NutriBites rely on querying and aggregating the historical nutritional data logged or scanned by the user.

Accuracy & Limitations

Running our own food database changed what we can promise here. When a scan resolves against a record we have verified ourselves, the nutrition values come from that record rather than from a generic estimate, which is where most of the error in photo-based tracking used to come from.

• Meal Scanning: NutriScan identifies food at 95% or better accuracy for items covered by our database. This is our own figure, based on our internal food-data verification, not a third-party certification.
• Manual Logging: Manually logged items resolve directly against the same verified records, so they are typically more accurate again than a scan, because there is no image interpretation step.

Two honest caveats we will not paper over. Portion size is the hard part: identifying the dish correctly is not the same as judging how much of it is on the plate, and depth, angle and plating all affect that estimate. Coverage sets the ceiling: accuracy is highest for foods already in our database, and a dish we have never seen falls back to search and AI estimation, where the error is wider. Both are why every result stays editable, and correcting one teaches the system.

We are continuously working to improve our algorithms and data validation processes to minimize these errors.

By integrating these diverse and high-quality sources, NutriScan aims to deliver trustworthy, comprehensive nutritional insights for all types of food consumption, helping you make informed dietary choices and effectively track your progress towards your health goals.