VesselRef
Answers from scanned manuals and drawings, without leaving your network.
Ordinary document AI reads text. Technical documentation is scans, drawings and tables — there is no text to read, which is why normal tools return nothing useful. VesselRef retrieves on the page as an image and extracts tables cell by cell, with human review, and runs entirely on the customer's own machine. Built end to end: visual-page retrieval, local generation, legacy Office and Outlook parsing, and a deployable web application with authentication.
What this proves
That I can build genuinely hard technology alone — model selection, retrieval architecture, local inference, messy real-world file formats and a deployable authenticated application, not a wrapper around somebody's API.
What it does
- ✓ Visual-page retrieval (ColQwen2) — works on scans and drawings
- ✓ Local generation; nothing leaves the customer's network
- ✓ Cell-level table extraction with human review
- ✓ Legacy Office and Outlook format parsing
- ✓ Deployable web app with authentication
Built with
My role
Solo — architecture, retrieval, extraction pipeline, application and deployment.