This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend.
What I Built
I built AllergenPal, a smart, local meal planner designed specifically for my roommate, Sarah, who manages severe, combined allergies to peanuts, dairy, and gluten. Planning daily meals that safely avoid all three without getting repetitive is an ongoing challenge for her.
AllergenPal acts as a dedicated culinary companion. It takes whatever ingredients are currently in our fridge, references Sarah's strict allergen profile, and dynamically generates completely safe, delicious, and balanced recipe ideas for the week.
Demo
• [https://allergen-pal-kappa.vercel.app/](https://allergen-pal-kappa.vercel.app/)
• [https://github.com/VishwajeetCSE/AllergenPal](https://github.com/VishwajeetCSE/AllergenPal)
How I Built It
The application is built around Gemma, Google's open-weight model.
• Local Inference: I run Gemma locally on my laptop using Ollama to handle all the natural language recipe generation.
• Architecture: The front-end is a lightweight interface where you toggle specific allergen profiles and list available ingredients. This data is structured into a rigorous system prompt that forces the local Gemma instance to parse recipes, strip out high-risk components, and substitute them with safe alternatives (like using coconut milk instead of dairy).
Why Does Open Innovation Matter?
Open innovation made this project entirely possible and practical for three major reasons:
Privacy of Personal Health Data: Dietary restrictions and health vulnerabilities are deeply personal. By running Gemma locally, Sarah's medical and lifestyle data never leaves our local network or gets processed by external corporate logging servers.
Zero Operating Costs: Closed-source APIs charge per token, making iterative brainstorming sessions for weekly meals expensive over time. A local open-weight model costs absolutely nothing to query indefinitely.
True Local Customization: Using an open model allowed me to tightly constrain the system prompt parameters directly on my hardware, ensuring highly deterministic, allergy-safe outputs without background corporate alignment filters interfering with the ingredient logic. Prize Categories • Best Use of Gemma: Used Google's open-weight model as the foundation of the project's local intelligence layer.