Table Snap Pro
Web app that turns photos of tables into structured data and ready-to-use Excel files, powered by multimodal AI models through OpenRouter.
Overview
Table Snap Pro automates a task that is usually done by hand: transcribing tabular data from screenshots, scans, or photographs into a usable digital format. You upload an image, the app detects the table, and you get structured data you can preview and download as an Excel file.
The problem
Transcribing tables from images by hand is slow and error prone, and hard-coding a single AI model makes any service built this way fragile: free vision models appear, hit rate limits, and disappear without notice.
What I built
I built a Next.js application whose extraction endpoint streams progress over Server-Sent Events while it discovers, ranks and attempts vision models until one returns valid data. Every step of the pipeline is reported to the interface: catalog lookup, filtering, model attempts, parsing and completion.
Technical decisions
- Dynamic model discovery instead of a fixed model, so the service keeps working when individual models become unavailable.
- A strict output contract in the prompt (valid JSON only, row headers first, equal-length arrays) to make parsing deterministic.
- Server-Sent Events for step-by-step progress, keeping long extractions transparent to the user.
- SheetJS on the client side to turn structured data into a downloadable spreadsheet.
Stack
- Next.js
- React
- TypeScript
- Tailwind CSS
- OpenRouter API
- SheetJS