Offline menu description writer for small restaurants with Gemma 3 4B

What they built

Many small restaurants and tea shops in Batticaloa list only a dish name and a price, so visitors may not know what the food is. Writing descriptions takes time or money, and cloud AI tools would mean sending the shop's menu and prices to a server the owner doesn't control.

The author built a small Python and Streamlit app for those owners. They enter the dish name in English and Tamil, add a note such as "spicy, fresh crab", and click one button; Gemma 3 4B running locally through Ollama writes a short, appetizing English description on an ordinary laptop with no internet access.

It is a working prototype tested on dishes like Chettinad Chicken, Idli and Filter Coffee, built for the Hacktoberfest Open-Source AI Challenge. The author has not yet tested it with a real shop owner.

Setup

• Gemma 3 4B (gemma3:4b) running locally through Ollama
• Python with Streamlit for a one-button page
• A normal laptop, offline once the model is downloaded

How it works

1. The prompt asks for one appetizing English description, two sentences at most, with no extra text.
2. The model is set in one line of code, so other open models (Llama, Qwen) can be swapped in.

Results & numbers

• A description in about 20 seconds per dish
• Good English output from the 4B model

Gotchas

• The 4B model's Tamil was unreliable: one Idli result was meaningless text and another mixed in broken transliteration
• Design fix: the owner types the Tamil name and the model writes only the English description

“A small model has limits, and designing around them made the product more honest and useful.”— ajaypraneeth15

Do it yourself

1. Install Ollama and run ollama pull gemma3:4b
2. pip install ollama streamlit
3. python -m streamlit run app.py
4. Enter the dish name in English and Tamil plus a short note, then generate

ollama pull gemma3:4b
pip install ollama streamlit
python -m streamlit run app.py