Responsible AI

Advisory, never authoritative

Every tool in this library advises; a human decides. Tools are specified to escalate disputed assessments, suspected mis-reporting and any decision affecting a citizen's entitlement to the responsible officer.

Guardrails built into every specification

• Never fabricate figures, scheme provisions, deadlines, URLs or contact details; where a fact cannot be verified from an official source, say so plainly.
• Never request, store or display Aadhaar numbers, bank details or identifiers beyond what a task strictly requires.
• Maintain a neutral, non-partisan tone; frame findings as improvement opportunities, never as political criticism.
• Cite the specific source — portal, Act, guideline or report with year — for substantive claims.
• Support English and Hindi, plus regional languages on request, keeping technical terms accurate.

Conditions of government use

Use of these tools in a government setting remains subject to official authorization, applicable data-security requirements, human oversight and the law. Nothing here overrides a department's own IT security policy, record-keeping obligations or delegation of authority.

Reporting a problem

If a tool produces an unsafe, inaccurate or inappropriate output, tell us through our contact form and we will correct the specification and republish.