Landing Page
Ingredient Identifier

Smart Vegan Recipes – Build Story

Elevator Pitch

Smart Vegan Recipes is your AI sous‑chef: snap a photo of the produce on your counter and, in seconds, the app recognizes each item and serves up a plant‑based, nutritionally balanced recipe tailored to the cuisine you crave—reducing food waste and decision fatigue with instant, kitchen‑ready inspiration.

Inspiration

My wife is an avid gardener whose harvest overflows with colorful vegetables and greens. While the abundance is a blessing, it often leaves us asking the same nightly question: “What can we possibly cook with all of this?” I wanted a tool that would celebrate her hard work, cut down on waste, and make choosing dinner as easy as taking a photo.

What It Does

Snap & Detect – Users take a picture of their fresh ingredients.
Visual Recognition – The app identifies each item and displays a list of detected produce.
Cuisine Picker – Select from global cuisines (e.g., Italian, Thai, Mexican).
AI Recipe Generation – A bespoke vegan recipe appears with step‑by‑step guidance and nutrition facts.
Save & Share – One‑tap saving to a personal cookbook and social‑media ready share cards.

How We Built It

Layer	Tech & Approach
Mobile Front End	React  TypeScript, and Tailwind‑like styling for clean, scalable UI.
Computer Vision	Gemini AI
Recipe Engine	Gemini AI GPT‑4 function‑calling pipeline that maps recognized ingredients to cuisine templates and USDA nutritional data.

Challenges We Ran Into

Ingredient Ambiguity – Kale varieties and similar leafy greens confused the vision model; we added confidence thresholds and “confirm/edit” chips.
Ingredients Confusion - A picture of my wife with Cabbage and Tomato Plants showed both as ingredients and only gave recipes with Cabbage and Tomato Plants.
Prompt Grounding – Early LLM prompts produced non‑vegan recipes (whoops!); explicit system messages and strict ingredient lists fixed this.
Performance – Image uploads on rural 3G tested patience; introduced local compression and progressive loading.
Cost Control – Vision API calls add up; batching images and caching common produce reduced spend by ~40 %.

Accomplishments That We’re Proud Of

95 % detection accuracy on a 20‑item produce test set.
End‑to‑end recipe generation averages 20 seconds on 4G.
Under an hour from idea to reality.

What We Learned

Simplicity > Features – A single camera button outperforms multi‑step ingredient pickers.
Contextual Prompts beat generic ones for LLM consistency.
Gardeners love seeing their veggies spotlighted—personal connection drives usage.

What’s Next for Smart Vegan Recipes

Voice‑Only Cooking – Hands‑free, smart‑speaker integration.
Smart Shopping List – Suggest pantry staples to complement garden produce.
Premium Tier – Advanced nutrition tracking and chef‑curated recipe packs and even food delivery once user traction validates monetization.

Together, we’re turning backyard bounty into effortless, inspired vegan meals—one snapshot at a time.