This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

RoomiePlate is an allergy-aware weeknight meal planner for the person you share a kitchen with.

The friend in this build is Alex: peanut and shellfish allergies, vegetarian-friendly dinners when possible, and a hard 30-minute weeknight budget. You set the roommate profile (allergies, dislikes, cuisines, pantry), generate a week of dinners, swap or regenerate a day, and export one grocery list that keeps the allergen note at the top.

Every meal card shows a safety line the app writes after a scan — for example SAFE for Alex: no peanut/shellfish — not whatever the model claimed.

It is a planning aid, not medical advice. Labels still matter, and a shared kitchen can still cross-contact a “safe” recipe.

Demo

Live demo: https://roomieplate.vercel.app (uses offline fixtures when no local Ollama is present)
Source: https://github.com/icohangar-ops/roomieplate

ollama pull gemma2:2b
ollama serve
git clone https://github.com/icohangar-ops/roomieplate.git
cd roomieplate
npm install
npm run dev

Open http://127.0.0.1:43123, keep Alex’s profile, hit Plan Alex week. The header pill should say Local gemma2:2b when Ollama is up.

Screenshots and a click-through live in docs/ in the repo.

Link to Code

icohangar-ops / roomieplate

RoomiePlate — allergy-aware meal planner for a roommate (local Gemma/Ollama). Hacktoberfest Weekend Challenge 2026.

RoomiePlate

RoomiePlate is a weeknight meal planner for the person you live with. It builds a week of meals that hard-avoid their allergies, then turns the plan into one grocery list.

The friend in this project is Alex: peanut and shellfish allergies, a preference for vegetarian dinners, and a 30-minute cap on weeknights. Those details are fiction for the demo. The rule is not: if an allergen is on the profile, it does not get a seat at the table.

This repo was started for the Hacktoberfest Weekend Challenge, October 2–5, 2026.

Who it is for

Shared kitchens fail in small, specific ways. Someone buys the wrong sauce, a “quick noodle” recipe hides peanut butter in step two, or Thursday’s dinner gets planned by whoever is hungriest. RoomiePlate is for the roommate who cooks and wants the constraint written down before the pan is hot.

It is a…

View on GitHub

How I Built It

Next.js + TypeScript + Tailwind for the planner UI
Ollama OpenAI-compatible API on 127.0.0.1:11434
Gemma 2 (gemma2:2b) as the default open-weight planner (swap with OLLAMA_MODEL)
A strict system prompt plus a post-generation allergen scanner that throws out any meal whose title, ingredients, or steps mention a profile allergen
Offline fixtures so demos and deploys still work without a GPU — clearly labeled in the UI

The product path is local Gemma. Fixtures are the backup, not the pitch.

I built this during the challenge window (Oct 2–5, 2026) as a brand-new repo — no recycled PR hunt.

Why Open Innovation Matters

Allergy notes look like grocery text, but they are health-adjacent personal data.

Stay local: with Ollama + Gemma, the dietary profile goes to a process on your machine, not a hosted LLM vendor.
Swap freely: change one env var to try another open model without rewriting the app.
Trust, then verify: small models slip. RoomiePlate does not trust prose — it rescans and replaces unsafe meals before they hit the board.
Cheap / offline-capable: after the model pull, planning does not need a cloud API bill.

For a tool that decides whether peanut butter “sneaks into step two,” open and local fit better than shipping that profile to a closed API by default.