Tablespeak: a menu translator that spells every dish the way the restaurant does

Description and use

Tablespeak enables restaurant owners to generate accurate, multilingual menus with consistent dish names and descriptions, leveraging open‑weight LLMs and a lightweight web interface.

Problem

Restaurant owners need accurate, consistent menu translations without paying for each edit, especially for dishes with unique local names.

Outcome

Tablespeak delivers accurate, consistent multilingual menus with minimal effort, reducing translation costs to fractions of a cent and eliminating the need for manual edits.

Workflow

• 1. Owner pastes menu text (Gujarati, Hindi, English, or mix) into the web app.
• 2. Owner enters a glossary line per dish in the format "Gujarati / transliteration | English | description".
• 3. Owner selects target language from a dropdown.
• 4. Browser sends menu and glossary to a Node server.
• 5. Server builds a prompt and calls an OpenAI‑compatible API (e.g., Qwen 2.5 72B, Llama 3.3 70B, Gemma 3 27B via OpenRouter).
• 6. Server returns a formatted menu card with consistent spellings and a turmeric dot for glossary items.
• 7. Owner reviews, prints, or shares the card.

Evidence

The part someone would pay for is the glossary. A generic translator guesses dish names. It turns ઊંધિયું into "mixed vegetable curry" and spells khaman three different ways on one menu.

Source

The whole project has zero dependencies: plain HTML, CSS and JavaScript, plus about 100 lines of Node.

Source

I used Qwen 2.5 72B, Llama 3.3 70B and Gemma 3 27B through OpenRouter, and switching between them is a dropdown in the app and one line in the code.

Source

Free models disappear. Two of my three free model ids started returning 404 partway through, with an error saying the free version was gone.

Source