What this is, what works, and what is pretend

Anyone whose grievance was closed while their problem continued. In 2026 DARPG published an analysis of recurring complaints and found 5.01 lakh grievances across ten categories and six ministries that were the same problems coming back between 2022 and 2025. The single largest category was PM-KISAN instalments stopping, at 1,52,102 grievances.

This prototype answers a real telephone. Dial the number, describe a problem the way you would describe it to a person, and a case is filed while you are still on the call. The case number is read back to you twice before you hang up.

Afterwards, sign in to the demo, click “Filed it by phone?” and type the number the agent read out. The case moves onto your account and behaves like every other case from there — success checks, the department's claim, your verdict, the appeal draft.

Sarvam AI does the listening and the speaking. The understanding — what the complaint actually is, which department owns it, and what would have to be true for it to count as fixed — runs in this project's own edge function, writing to the same database the website uses. The audio is never stored. Only the text is.

The portal is fast and its numbers are good — central ministries disposed of 1,88,969 grievances in April 2026, averaging 13 days. The difficulty is what those numbers measure.

A grievance is recorded as disposed when the department writes its reply, not when the problem ends. And a closed grievance cannot be reopened — the FAQ asks the citizen to file a fresh one. So a problem that survives closure comes back as a new grievance, gets a new number, and is counted again as a new receipt and a new disposal.

There is an appeal channel above this, heard by an officer of Additional or Joint Secretary rank, and it works — it disposes 93.84% of what reaches it. Very little reaches it. It has to be filed within 30 days of a closure nobody tells you how to challenge.

• “Resolved” is given something to be checked against. Every case carries what has to be true for it to count as fixed — the amount credited, the supply running, the document in hand — derived from the kind of problem it is.
• The department's claim is not the verdict. When a fix is claimed, you are asked the plain question rather than whether your grievance was redressed, with those conditions in front of you unanswered. Only you can settle them.
• History is carried forward, not reset. A repeat filing keeps every earlier registration number, closure date and remark, so the elapsed time is counted across the whole problem rather than restarting at zero.
• Deadlines are tracked and the letters are written. The appeal and RTI dates are computed from the case and shown before they lapse. The letters are drafted from stored facts and sent by the citizen, never automatically.

It needs no new government process. Every step it tracks already exists in published policy and statute — the 21-day redress standard, the 30-day appeal to the Nodal Appellate Authority, and the Right to Information Act. The machinery is built and running below capacity. What is missing is a citizen-side layer that knows the dates and prepares the paperwork.

To run for real it would need read access to a citizen's own grievance status and closure remarks, and a submission route into the existing appeal channel. Nothing here requires access to anyone else's data: the success checks, verdicts and history are the citizen's own record of their own case.

The phone line is the part that generalises furthest, because the people worst served by a portal are the ones least likely to open one — and it is also the part with the clearest regulatory line through it. A citizen dialling a published number is not telemarketing, so an inbound line needs a number with KYC and nothing more. That is why this one exists and answers today.

The reverse direction is where the wall is. Calling a citizen back, or texting them, means DLT registration as a telemarketer with a registered sender ID and approved templates — a compliance path, not a weekend build. Which is a pity, because the callback is the strongest thing this design does not have yet: when a department files its action-taken report, the case already knows the citizen's number and already knows what would count as fixed, so the right move is to ring them and ask. CPGRAMS calls a sample of citizens to ask that very question, so this would be an upgrade to something the department already does rather than a new process. The escalation engine does not care which transport carries it.

• No live government system is contacted, tested or interfered with.
• No real Aadhaar, PAN, OTP, payment or health data is used anywhere.
• No government emblem is used, and nothing claims official approval.
• The appeal and RTI drafts restate facts already on the case file. They are a starting point for a citizen to edit, not legal advice.