VISION (previously called gemini-research) is a multimodal workspace I built around the Gemini API. The main idea was to see how well an AI could handle visual tasks like turning hand-drawn sketches into working code, or extracting structured data from messy documents.
- Sketch to Code: You can upload a wireframe or a rough sketch, and it will generate the HTML and CSS for it. There is a strict mode for exact copies and a creative mode where the AI takes some design liberties.
- OCR and Data Extraction: It reads handwritten or printed text and can parse things like invoices or tables into clean JSON data.
- Interface: I built the UI to look like a clean, floating dark-mode IDE using CSS glassmorphism.
- Run npm install to get the dependencies.
- Create a .env file and add your GEMINI_API_KEY. You can also specify a PORT (defaults to 3000).
- Start the server with node server.js.
- Open http://localhost:3000 in your browser.