Digital Archive · School of Sciences, UCV

Context

The archive of the School of Sciences at Universidad Central de Venezuela holds personnel files and academic records on paper, with decades of history and no budget for commercial software. Built with Susana Carvallo.

The problem

Retrieving a file could take days. And any AI solution carried two risks that are unacceptable in a public institution: that API spend spirals out of control, and that the assistant modifies or deletes records on its own.

What I did

• Document management platform with metadata search, audit trail, trash, backup and strict separation between Archive and HR
• The API key never leaves the server: the browser talks to the backend, and the backend calls the model
• The user role decides which tools exist. There are three modes — public, read-only and editor — each with a different set of available tools. The prompt is the first line of defence and the weakest; the real barrier is the tool not existing at all
• The AI proposes and a human approves. No change is written without someone pressing a button
• Deliberate prompt caching: nothing volatile enters the cached block, so context writes are not paid for on every turn
• A plain-HTTP client against the provider, with no extra dependencies, to get the real cost of each call instead of estimating it
• A scanning application to digitise the physical files

Outcome

System in production, public and under active development. Still to be published: files digitised, lookup time before and after, and monthly AI cost.

Stack