Allergy & Diet Guard — Protecting My Friend with Local Open-Source AI

Dien — Fri, 02 Oct 2026 11:20:52 +0000

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

What I Built

I built Allergy & Diet Guard (SafeBite) for my friend Sarah, who lives with an unforgiving combination of dietary hazards:

Severe Peanut Allergy (life-threatening anaphylactic shock)
Celiac Disease (strict gluten-free requirement)
Lactose Intolerance (no dairy)

Eating out at local restaurants, street food stalls, or night markets with Sarah is usually stressful. Menus and product labels rarely say "Peanut" or "Wheat"; they list obscure culinary terms like "arachis oil", "seitan", "malt extract", "casein", or "modified starch". A single contaminated bite can mean an emergency hospital visit.

Allergy & Diet Guard is a local, privacy-first food safety sentinel. It takes any food description, ingredients list, or menu item and runs a dual-layer open-source safety analysis to:

Detect both direct allergens and obscure hidden derivatives (e.g. flagging arachis oil as a peanut risk).
Highlight high-risk kitchen cross-contamination vectors (shared fryers, woks, cutting boards).
Issue an unmistakable verdict: 🔴 DANGER, 🟡 WARNING, or 🟢 SAFE.
Generate a one-click, bilingual (English & Indonesian) Chef Safety Card that Sarah can show directly to the waiter or kitchen staff.

When I showed the prototype to Sarah, she breathed a huge sigh of relief:

"Finally, something I can use in the basement food court without needing an internet signal or wondering if my medical details are being tracked by an ad network!"

Demo

Live Web App (Hosted on Render): https://hacktoberfest-2026.onrender.com/weekend/
Unified Challenge Portal: https://hacktoberfest-2026.onrender.com/

The app runs as a zero-dependency lightweight web dashboard, deployed live as a Python web service on Render, or runnable locally at http://localhost:8080.

Real-world Test Scenarios:

Test 1: Pad Thai with hidden Peanut oil
- Input: "Pad thai noodles with fried egg, bean sprouts, crushed peanuts, scallions, cooked with arachis oil and fish sauce."
- Verdict: 🔴 DANGER: DO NOT CONSUME
- Detected Hazards: Direct peanuts + obscure alias arachis oil.
- Kitchen Risks: Shared woks and cold-pressed peanut oils.
Test 2: Vegan Stew with hidden Gluten
- Input: "Vegetarian plant-based BBQ bowl with roasted seitan chunks, malt extract sauce, steamed corn, and barley pearls."
- Verdict: 🔴 DANGER: Celiac / Gluten Violation
- Detected Hazards: seitan, malt extract, and barley pearls.
Test 3: Clean Atlantic Salmon
- Input: "Pan-seared Atlantic salmon fillet with extra virgin olive oil, sea salt, cracked black pepper, steamed broccoli, and jasmine rice."
- Verdict: 🟢 SAFE
Chef Safety Card Output (English & Indonesian):

⚠️ ALLERGY NOTICE FOR THE CHEF:
Hello, I am ordering for my friend Sarah.
STRICT ALLERGIES: Peanut / Groundnut, Gluten / Wheat / Celiac Risk, Dairy / Milk / Casein
DIETARY PREFERENCES: None
Please ensure food contains NO traces of these ingredients and clean cookware/utensils are used to avoid cross-contamination. Thank you!

Code

The code is completely open-source and structured for immediate zero-dependency execution:

Repository: github.com/dieennn/hacktoberfest-2026/tree/main/00-weekend-allergy-guard

Project Architecture:

app.py: Standalone Python web server & REST API using purely Python 3 standard library (http.server & socketserver). Zero pip install required.
analyzer.py: Open clinical taxonomy engine mapping FDA Big 9 and EU 14 allergens, cross-contamination rules, and local inference bridge.
static/index.html: Responsive, accessible web UI with friend profile presets and instant safety feedback.
test_analyzer.py: Self-contained assert-based automated test suite.

How I Built It

The core engine is built on a hybrid open-source AI architecture:

Embedded Clinical Semantic Ontology (100% Offline & Deterministic): We codified FDA Big 9 and EU 14 allergens into an open clinical knowledge base with dozens of obscure culinary aliases (e.g. arachis, valencias, seitan, farro, spelt, caseinate, surimi, albumen) and cross-contact risk patterns.
Local Open-Weights LLM Integration (Ollama): When Ollama is running locally (llama3.2, gemma2, or qwen2.5), the system passes the matched hazards to the local open model to generate tailored culinary advice and allergen-free ingredient swaps.
Failsafe Design Ladder: Safety cannot rely on probabilistic guesswork. If either the deterministic clinical rules or the local open model identifies a risk, the system escalates to DANGER.

Why Does Open Innovation Matter?

Why not just use a closed proprietary API like ChatGPT or Claude?

Works Completely Offline in Low-Connectivity Environments: Food courts, basement restaurants, rural markets, and international travel frequently suffer from poor or zero internet connectivity. Closed APIs fail completely. Open-source local inference works everywhere, every time.
Health Data Privacy & Sovereignty: A person's medical allergies and health conditions are strictly sensitive personal data. Running open-source models locally guarantees zero data telemetry and zero tracking.
Zero Cost & Zero Rate Limits: Friends shouldn't have to pay subscription fees or worry about API token credits just to know if their dinner is safe to eat.
Deterministic Failsafe Control: Closed AI models change behind closed doors and frequently suffer from hallucinations on complex ingredient derivatives. With open code and local weights, we enforce strict verification and safety boundaries.

My Agent Session

This project was built and validated during our Hacktoberfest 2026 preparation session using the DevRelay harness.

Prize Categories

Hacktoberfest Open-Source AI Challenge
Build for a Friend Theme
Best Use of Render: Deployed the live unified AI food safety application as a cloud web service on Render, serving both the interactive UI and the real-time inference API.

DEV Community: Dien