Petugram

Inspiration

We kept running into the same small, quiet failure over and over: food going bad in the back of the fridge before anyone remembered it was there. Not from carelessness — just because the fridge is invisible. You don't see what you have, so you buy duplicates, forget what's expiring, and eventually throw out something that was still good. At the same time, when there was surplus food, there was no easy way to get it to someone nearby who could actually use it. Food discovery — restaurants, recipes, food creators — was scattered across five different apps. We wanted one place that closed that whole loop: see what you have, cook it, plan around it, share what you can't use, and discover more food culture along the way.

What it does

Petugram is a zero-waste kitchen companion and social food platform built around one core loop: Manage → Cook → Plan → Share → Discover → Create → Connect → Impact.

Digital Fridge — scan a receipt, photo, or barcode (or add manually) to track ingredients and expiry dates, with alerts before things spoil. AI Match — generates recipes from whatever's actually in your fridge, ranked by fit, with substitutes suggested for missing ingredients. Meal Plan — builds a 3–7 day plan personalized to cuisine, time, and diet, and auto-generates a grocery list of only what's missing. Share + Inbox — post surplus food for nearby users to claim, then chat directly to arrange pickup. Discover — find restaurants, dishes, and recipes on an interactive map with rating and dietary filters. Social — posts, reels, and stories for sharing what you cooked, following other home cooks and creators. Impact — tracks food saved, money saved, and CO₂ avoided, localized to your region. How we built it

Petugram runs on a three-tier architecture: a vanilla JavaScript single-page frontend, a FastAPI (Python) backend, and a data/services layer combining MongoDB with AI services and external APIs.

Frontend: HTML5, CSS3, vanilla JavaScript Backend: Python, FastAPI, Uvicorn Database: MongoDB via PyMongo, with dedicated collections for Users, Fridges, FoodItems, Recipes, MealPlans, GroceryLists, SharePosts, Claims, Messages, Posts/Reels/Stories, Restaurants, Reviews, Achievements, and Impact AI/ML: Hugging Face inference for recipe generation (T5-based) and vision models for food/receipt recognition, with sentence-transformer (MiniLM) embeddings used to rank generated recipes against the ingredients a user actually has External APIs: Open Food Facts for barcode/product lookup, OpenStreetMap + Overpass + Nominatim for restaurant discovery and geocoding

The AI Match flow is a good example of how the pieces connect end-to-end: the frontend requests fridge data, FastAPI pulls it from MongoDB, forwards the ingredients to the AI service, gets recipes back, ranks and filters them, and returns results to the user — all in one request cycle.

Challenges we ran into

Ranking recipe suggestions in a way that actually felt relevant was harder than generating them — a raw recipe-generation model happily proposes dishes using ingredients you don't have, so we had to layer sentence-transformer embeddings on top to score and filter by real fit against the fridge inventory. Tying together five distinct feature areas (fridge, planning, sharing, discovery, social) into one coherent data model without it turning into five disconnected apps also took real iteration on the class/relationship design — a lot of the class diagram in our docs is the result of rethinking ownership boundaries between Users, SharePosts, Claims, and MealPlans more than once.

Accomplishments that we're proud of

We're proud that Petugram isn't just a concept — it's a fully working, deployed prototype covering the entire loop from scanning food into the fridge through to sharing surplus and viewing environmental impact. Every core feature described in our pitch — Digital Fridge, AI Match, Meal Plan, Share, Discover, Social, Impact — is live and functional in the deployed build, not mocked up.

What we learned

Building the AI layer taught us that model choice matters less than the surrounding system — recipe generation alone wasn't useful until we added ranking, filtering, and substitution logic around it. We also learned a lot about designing for a two-sided sharing feature (surplus food offers and claims): it's not just a CRUD flow, it needs real-time-feeling coordination (notifications, inbox, status tracking) to actually work for two strangers arranging a handoff.

What's next for Petugram Mobile app — moving from web prototype to native mobile Better AI / our own models — reducing dependence on external inference APIs Bigger food and recipe database — broader cuisine and ingredient coverage Personalized recommendations — using cook history and saved items to improve Discover Community expansion — more cities → more users → more food sharing → greater impact Partnerships — restaurants, grocery stores, food brands, and sustainability organizations, plus a creator ecosystem connecting creators, restaurants, brands, and users