This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
The Friend: Priya
The Problem: Priya is a busy freelancer who loves trying out new recipes, but she constantly struggles with meal planning around her fluctuating schedule, her sudden pantry cravings, and her specific dietary preferences (she loves vegetarian fusion and is trying to cut back on dairy). Manually searching for recipes and matching them to what's already in her fridge takes up way too much time.
The Solution: I built PantryChef AI, a smart, local-first recipe assistant tailored specifically for Priya. She can snap a quick photo of her fridge or type out what ingredients she has left, and the app instantly generates personalized, healthy fusion recipes with exact step-by-step instructions.
Demo
Live App: https://pantrychef-ai.example.com
Video Walkthrough: Watch the 2-minute demo on Loom (Bonus: She tested it live with leftover bell peppers and paneer, and loved the resulting Makhani tacos!)
Code
Python
Core recipe generation logic using a local open-weight model
def generate_custom_recipe(ingredients, preferences):
prompt = f"Create a quick vegetarian, low-dairy recipe for Priya using: {ingredients}."
response = local_llm.inference(prompt, temperature=0.7)
return format_recipe_card(response)
GitHub Repository: github.com/username/pantrychef-ai
How I Built It
Tech Stack: React (Frontend), Python/FastAPI (Backend)
Open-Source AI / Models: Powered locally by Gemma 2 (2B / 9B open-weights) via Ollama, ensuring zero latency and complete privacy for her grocery lists and personal preferences.
Architecture: The React frontend captures ingredient inputs and sends them to a lightweight local FastAPI backend, which prompts the Gemma model running locally to curate and stream back structured JSON recipe cards.
Why Does Open Innovation Matter?
Using open-source tools and open-weight models like Gemma made this project possible because Priya values her kitchen privacy and didn't want her personal dietary notes and fridge contents logged on a third-party commercial cloud server. Running everything locally via Ollama meant zero API costs, zero rate limits, and instant offline inference right from my laptop while testing it out in her kitchen.
My Agent Session
View my development agent session with DevRelay
Prize Categories
Best Use of Gemma
Best Use of a .tech Domain
Top comments (0)