Spaces:
Running
Running
| title: NutriLens | |
| emoji: 🔬🥗 | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: gradio | |
| sdk_version: "6.16.0" | |
| python_version: "3.11" | |
| app_file: app.py | |
| pinned: true | |
| license: apache-2.0 | |
| tags: | |
| - food | |
| - nutrition | |
| - health | |
| - science | |
| - hackathon | |
| - build-small | |
| - track:backyard | |
| # NutriLens: Food Health Impact Analyzer | |
| Snap a photo of your meal, grocery label, or type a list of ingredients. | |
| NutriLens identifies each ingredient, looks up real nutritional data from | |
| the USDA, finds relevant scientific studies on PubMed, and delivers a clear | |
| per-ingredient health breakdown with proper citations. | |
| Works with food labels in **any language**. | |
| ## Why this exists | |
| Most of us buy food every day without really knowing what's in it or what | |
| it does to our bodies. Ingredient lists are full of names that sound | |
| foreign even in our own language - "sorbitol syrup," "soya lecithin," | |
| "emulsifier" - and nutrition science lives in dense academic papers most | |
| people will never read. | |
| That gap matters: the foods we eat regularly shape our long-term health, | |
| and a lot of that influence is invisible until it's added up over years. | |
| NutriLens exists to close that gap - to take what's already known and | |
| published and turn it into something anyone can read in a minute, in | |
| plain language, before they decide what to put in their cart or on their | |
| plate. | |
| The goal isn't to scare anyone away from a treat or declare foods "good" | |
| or "bad." It's awareness: knowing what you're consuming, what the science | |
| actually says about it, and why - so you can make your own informed | |
| choices. | |
| ## How it works | |
| 1. **Identify**: A small vision-language model reads your food photo or label | |
| and extracts the ingredients. | |
| 2. **Look up**: Each ingredient is matched against the USDA FoodData Central | |
| database for verified nutritional data. | |
| 3. **Research**: PubMed is searched for recent scientific reviews on each | |
| ingredient's health effects. | |
| 4. **Analyze**: The model synthesizes the nutritional data and study findings | |
| into a clear, evidence-based health report with citations. | |
| When databases are rate-limited, the model falls back to its own knowledge | |
| and clearly labels those sections. | |
| ## Health focus areas | |
| General, Heart health, Anti-inflammatory, Blood sugar, | |
| Gut health, Energy, Bone health. | |
| ## Data sources | |
| - USDA FoodData Central (400K+ foods) | |
| - PubMed / NCBI E-utilities (peer-reviewed literature) | |
| ## Built for | |
| [Gradio Build Small Hackathon](https://huggingface.co/build-small-hackathon) (June 2026) | |
| ## Demo video | |
| [Watch the demo](https://youtu.be/ZAhTMsoH_n8) | |
| ## Social post | |
| [See the announcement post](https://x.com/i/status/2066630127283318851) | |
| ## Team | |
| - [@vicarioush](https://huggingface.co/vicarioush) | |