Driving legacy screens and portals that offer no API

The last mile of an Indian back office is often a screen with nothing behind it you can call — a filing portal, a bank's corporate console, a state department form. A model reads the rendered page or the desktop, decides where to click and what to type, and performs the steps a clerk would, recording the run so somebody can replay what it actually did.

Effort
More than a few weeks
Skill level
Some technical skill
Organisation size
Mid-market
Value
Time saved, Cost saved

Tools named for this

  • A browser-automation agent that reads the page's accessibility tree rather than only pixels
  • A recorded run — screenshots plus the action list — kept as the audit artefact
  • A deterministic script for the steps that never change, with the model reserved for the ones that do

What to check before you ship it in India

  • Screen recordings of a filing or banking portal capture identifiers, account numbers and full names in the clear, and the recording is a new store of personal data you created and control. Section 8(5) puts the duty of reasonable security safeguards to prevent a personal data breach on the fiduciary, and run artefacts kept for debugging routinely outlive the purpose that justified them.
  • Evaluation on real desktops names the two standing weaknesses as locating the right element on screen and knowing the application's own conventions. A portal that renames a field is not a broken selector to this kind of system; it is a slightly harder guess, and it will make one.

Sources

Every claim on this page traces to one of these, on the date it was read.