All sectors

Data entry agent for business software without an API

An agent drives the interface of legacy business software as an operator would: it fills in the screens, generates the documents and flags missing data before confirming.

Talk to an engineer

Illustrative mock-up: the names, values and documents shown are fictitious.

Context

Management software with no usable API, where a team re-keys every file screen by screen.

The problem

Hours of manual data entry and transcription errors, with no possibility of API integration.

What we built

A computer-use agent on a virtual machine: a vision model plans the actions, a grounding stage locates interface elements, and a workflow engine breaks the journey down into tasks. A grounding benchmark of 40 pixel-annotated elements to choose the clicking strategy.

Steps

01Mapping of screens and the journey (11 steps)
02Grounding benchmark: general-purpose model, two-pass zoom, small specialised local model
03End-to-end runs on a test database
04Testing by the business user

Hosting and models

Client's Windows virtual machine; Gemini 3 Flash via Vertex AI; alternative tested: quantised Holo-3.1-4B grounding model run locally.

Services involved

Forward Deployed Engineers · Agentic platforms

We name a client only with their written agreement. The budgets and detailed results of our engagements remain confidential.