AiSewak
For representatives' public grievance lines

AI citizen helpline for elected representatives

Publishing a number is easy. Answering it at eleven at night, in the caller's own language, writing the complaint down accurately, sending it to an officer who will act on it, and ringing back when it is done — that is the part offices do not have the people for. An AI voice agent does the whole loop, and the last step is the one that changes anything.

Always answered
No busy tone, no queue, no office hours
Reference no.
Issued on the call and read back to the caller
Closed the loop
A callback when it is marked resolved, not before
Escalation matrix
Routed by category, jurisdiction and severity

A citizen helpline run by an elected representative's office is a different instrument from a government helpline. It has no statutory power: it cannot sanction, direct an officer, or bind a department to a timeline. What it has is standing — a department answers a representative's office faster than it answers an individual — and continuity, because the office is still there next month.

That distinction has to be built into the agent, not left to tone. A government line may state a service-level commitment; a representative's line may state only what it will do, which is record the matter accurately, push it to the right place, and tell the caller what came back. An agent that implies more than that damages the office the first time the implication fails.

AiSewak builds these lines for MLA, MP, mayor and corporator offices. The agent answers in the languages and registers of the area, files into whatever system the office and the administration already use, and runs the follow-up calling that a small office cannot staff.

The lifecycle, and where it usually breaks

Grievance systems are rarely lost at resolution. They are lost at intake and at closure — the two ends nobody is measured on.

  1. 1

    Intake that is actually usable

    The difference between water problem in Sector 4 and no supply on the 3rd-floor line of Block C since Tuesday morning, reported by three households, is the difference between a ticket an officer can act on and one that comes back for clarification. The agent probes until the description is actionable, then stops probing — an interrogation loses the caller as surely as a vague note loses the officer.

  2. 2

    Identification and reference number

    Caller name, a callback number confirmed digit by digit, and the location resolved to booth, ward or revenue village. A reference number is issued during the call and read back, because a caller with no reference has no way to follow up and will simply call again as a new complaint.

  3. 3

    Deduplication against what is open

    Before a new case is created the agent checks for an open matter at the same location in the same category. Several reporters on one case is information; several cases for one problem is noise that inflates the pending count and splits the assignment.

  4. 4

    Routing by jurisdiction, then department

    Which officer can fix this depends first on where the caller is standing. The agent resolves location to block, tehsil, ward or zone, then picks the department and the named post-holder, and writes the handover in the form that office accepts.

  5. 5

    Chase, on a schedule

    The unglamorous part. Scheduled follow-up calls to the department contact against each open reference, with each stated status recorded and dated. An office that can say we asked on the 4th, the 11th and the 18th is in a materially different position from one that can say we forwarded it.

  6. 6

    The closure call

    When a matter is marked resolved, the agent calls the person who raised it and asks whether it actually was. Confirmed closures are recorded; disputed ones reopen with the citizen's own words attached. Almost no office performs this step, and it is the only check on closure quality that exists.

What the line must never do

These are hard constraints in the agent's configuration, not guidance in a style document. Each of them is a way a public line loses the public.

Promise an outcome

The agent never says it will be fixed, never gives a date it does not control, and never implies the representative will intervene personally. It states what it has done and what it will do next.

Claim influence over a posting

Transfer and posting requests are recorded neutrally and passed on with no suggestion of leverage. The agent does not discuss any officer by name in that context.

Ask for money or credentials

No payment, no OTP, no UPI PIN, no bank or Aadhaar number beyond what a named scheme genuinely requires. The agent says so unprompted, because helpline impersonation is a live fraud pattern in India.

Take political positions

The line is for civic matters. It does not campaign, does not solicit support, does not comment on parties or candidates, and does not treat a caller differently by affiliation. An office line that behaves like a campaign line stops being called.

Handle an emergency itself

Fire, medical, violence, a child in danger, a woman in immediate danger: the agent breaks the flow, gives 112, 1098 or 181 as applicable, confirms the caller has it, and raises a flagged alert. It does not take a report and queue it.

Argue with a distressed caller

Abuse, threats and acute distress route to a human, with the transcript, rather than being managed by the agent. A line that traps an upset person in a loop produces exactly the story an office cannot afford.

Sitting alongside the statutory helplines, not on top of them

India already has a dense layer of official lines: 112 for emergencies, the state CM helpline or grievance portal, 1930 for cyber fraud, 14416 for mental health, 1098 for children, 181 for women in distress, plus department-specific numbers for electricity, railways and health schemes. A representative's line that tries to absorb all of this becomes a slower path to the same place.

The correct behaviour is a handoff with a reason. For anything with a designated statutory channel, the agent gives that number, explains what it does, and offers to register the matter with the office as well so that it can chase — which is the genuine added value, since the statutory line issues a ticket but nobody follows it for the citizen.

Where the state grievance portal exposes an interface, the agent files there and stores the returned reference against the caller, so the office and the administration are tracking the same ticket. Where it does not, the agent produces a complete case file that a staff member files in one action. Either way the office's register and the state's register point at the same thing, which is what makes a monthly closure report defensible.

What the office gets to see

Volume by category and by ward or booth, which almost immediately contradicts the office's intuition about what it is asked for most. Ageing of open matters against the department they sit with, so it is visible which department is the actual bottleneck rather than which is complained about most loudly.

Repeat clusters — the same street, the same category, for the third time — which distinguish a fix that did not hold from a new problem, and which are invisible in any system that closes and reopens tickets. And confirmed versus disputed closures, which is the single number that tells an office whether its reported resolution rate means anything.

The reporting is deliberately boring and defensible. Every figure traces to a transcript and a timestamp, so anything said publicly about the line can be substantiated. An office should not publish a resolution rate it cannot show the working for, and this is the reason to hold the working.

Frequently asked questions

What is an AI citizen helpline for an elected representative?

A published phone number, answered by a conversational AI agent, that takes citizens' grievances and requests for an MLA, MP, mayor or corporator's office. The agent answers at any hour in the caller's own language, records the matter in a form an officer can act on, issues a reference number, routes it to the correct department for that location, chases it on a schedule, and calls the citizen back when it is marked resolved to confirm that it was.

How is it different from the government's own grievance helpline?

Authority. A state CM helpline or department line can direct action within its own system and commit to a service level. A representative's office has no statutory power over a department — what it has is standing and persistence. So the agent is configured never to promise an outcome or a timeline, and to add value by chasing and by closing the loop with the citizen, which is the step statutory lines almost never perform. Where a statutory channel exists, the agent hands the citizen to it and registers the matter with the office to chase in parallel.

Can it file complaints on the state grievance portal automatically?

Where the state exposes an interface for it, yes, and the returned reference number is stored against the citizen and read back on the call. Where it does not, the agent produces a complete, correctly structured case file that a staff member submits in a single action rather than reconstructing from a phone note. Both paths keep the office's register and the administration's register pointing at the same ticket.

What stops it from being used as a campaign line?

Configuration, and it should be checked rather than assumed. The agent does not solicit support, does not discuss parties or candidates, does not vary its handling by a caller's affiliation, and does not collect voter-preference data on a grievance call. Mixing the two purposes is also a compliance problem: outbound political communication falls under the Election Commission's rules for registered political advertisers and under TRAI's TCCCPR, and a grievance line that drifts into campaigning inherits both.

What happens on an emergency or a distress call?

The agent stops the grievance flow immediately and gives the correct number — 112 for police, fire and ambulance, 1098 for a child in danger, 181 for a woman in distress, 1930 for cyber fraud, 14416 for mental health — confirms the caller has it, and raises a flagged alert to a human at the office rather than filing an ordinary ticket. Abuse, threats and acute distress are routed to a person with the transcript attached. The line is explicitly not an emergency service and never behaves as one.

Is an AI voice agent legal for outreach in India?

Yes, within a well-defined set of rules. The agent must disclose that it is an AI at the start of the call (ECI's 2024 advisory on synthetic media), outbound calling must respect TRAI TCCCPR 2018 and DLT registration, personal data must be handled under the DPDP Act 2023, and IT Rules 2021 govern the content itself. Election deployments additionally require registration with the ECI or the state CEO as a political advertiser. AiSewak ships these controls switched on by default rather than as an add-on.

Which Indian languages does the voice agent speak?

Hindi plus the 22 scheduled languages, and the regional variants that decide whether a call actually lands — Bhojpuri, Awadhi, Marwari, Mewari, Magahi, Santhali and others. Agents run on a mix of Indic-first models and the MeitY Bhashini stack, and a single agent can switch language mid-call when the person on the other end does.

How is the citizen's data protected?

Calls are recorded with the caller told at the start why, and the record carries a named purpose, a retention window and a deletion path under the DPDP Act 2023. Grievance data is not reused for campaign targeting, is not shared across deployments, and Indian data residency is available on request. A citizen can ask for their record to be deleted and the agent tells them how.

Publish a number your office can actually answer

Bring the ten complaints your front desk hears most and the officers each one has to reach. We will stand up a live line, run it for a week and show you the intake quality, the routing and the closure calls before anything is committed.

Every AiSewak agent identifies itself as an AI at the start of the call, never asks for an OTP or a payment, and honours DND. Election deployments require ECI / state CEO registration as a political advertiser.

AiSewak (AI Sewak) is a Voxdonna company, made in India.

© 2026 Donna AI Labs Private Limited · CIN U62013DL2026PTC464877. All rights reserved.