Menu-Sahayak：即時提供印度餐廳餐牌致敏原資訊（由 Llama 3.3 驅動）

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

Menu-Sahayak is a lightweight web app that analyzes Indian restaurant menus and flags likely allergens per dish.

Tourists with food allergies often struggle to read Hindi or regional-language menus. Generic translation tools will tell you what "Chole Bhature" means, but they won't tell you it contains gluten and dairy. Menu-Sahayak bridges that gap. Paste a menu, and instantly get dish names alongside clear allergen tags (dairy, nuts, gluten, eggs, soy) and a practical safety note for each item.

It solves a very real, high-stakes problem for travelers and locals alike, making dining out safer and stress-free.

Demo

Here is a quick look at the interface:
(Apna screenshot yahan upload karein — DEV editor mein image icon par click karke)

Code

Allergen Lens — Menu Allergen Checker for Indian Restaurants

A lightweight Gradio app that analyzes Indian restaurant menu items and flags likely allergens per dish.

What it does

Paste menu items (one dish per line), and the app returns a numbered list with:

dish name
likely allergens from: dairy, nuts, gluten, eggs, soy
one concise safety note

The model is instructed to avoid unsupported certainty: allergens are marked as likely/possible based on common preparation, and users should verify with restaurant staff for severe allergies.

Why this is useful

Tourists with allergies may struggle to read Hindi or regional-language menus. This tool provides quick, practical allergen guidance dish-by-dish.

Setup

Clone this repository.

Install dependencies:

pip install -r menu-sahayak/requirements.txt

Create env file:

cp menu-sahayak/.env.example menu-sahayak/.env

Add your Groq API key in menu-sahayak/.env:
```
GROQ_API_KEY=your_key_here
```
Run:
```
cd menu-sahayak
python app.py
```

Get a free Groq…

How I Built It

Model: Llama 3.3 70B (open-weight) served via Groq's fast inference API.
Frontend/UI: Python + Gradio (customized theme for a clean, mobile-friendly dashboard feel).
The Core Logic: The app streams the response from the model and uses a highly specific system prompt to force the model into an "Allergen Analyst" persona. Instead of creative writing, it outputs structured, numbered allergen lists.
Error Handling: Built-in graceful fallbacks for missing API keys, rate limits, and empty inputs.

Why Does Open Innovation Matter?

I chose Llama 3.3 70B (open-weight) over a closed API because:

Cost-Effective for a Side Hustle: Groq's generous free tier makes this app ₹0 to run. A closed model with similar reasoning capabilities would charge per token, eating into the thin margins of a side project.
Model Swap Freedom: Because the logic is built around an open-weight model, I am not locked into one provider. If I want to swap Llama for Qwen or Gemma tomorrow to improve allergy detection, it's a one-line change.
Transparency and Control: Open weights mean I know exactly what is processing the user's input. For an app dealing with health and dietary safety, this transparency is crucial.

The open pieces aren't just a buzzword here — the entire product is built around an accessible, open-weight model that I can control, prompt, and eventually self-host if needed.

Prize Categories

Overall Side Hustle Challenge Entry.