AI agents · Government & public offices
Scanned letters, WhatsApp messages and emails become structured cases with a due date, an owner and an escalation — so nobody has to ask what happened.
35+
AI agents running in our own production
A field the model was unsure of is reviewed by a person, never filed.
The everyday problem
A handwritten petition, filed in a register nobody reads
Forwarded to a department with no tracking number attached
The office learns it is late when the citizen returns
An office that acted looks unresponsive, because nothing was ever tracked.
What it does
Every channel lands in one queue, not five registers.
A handwritten Marathi letter becomes a structured case.
Every case carries a due date from day one.
The office is told before the deadline, not after.
The citizen can check status without visiting the office.
The month reads as sentences, not as a spreadsheet.
Ready to see it in action?
Send us ten scanned petitions — we will show you what the system reads out of them.
Or call +91 95525 86428
What changes
Today
Where the letter lives
An inward register
Who knows the status
Whoever forwarded it
When you learn it is late
When the citizen returns
The monthly report
Two days of counting
With the agents
Where the letter lives
A searchable case file
Who knows the status
The office and the citizen
When you learn it is late
The day before it breaches
The monthly report
Waiting on Monday morning
How the work moves
Scan the letter, or let the citizen send it.
Fields read, translated, categorised and filed as a case.
Assign an owner and route it to the department.
Reminders, overdue flags and escalation until it closes.
Record the resolution; the citizen sees it immediately.
Where it stops A field the model was unsure of is never filed silently — it goes to a review queue an operator clears before the case moves on. The agents register, track and remind; they do not answer a policy question or commit the office to an outcome. Every status change is recorded against a named person.
The deadline, drawn
The petition is scanned, extracted and registered with a number.
An owner is assigned and it is forwarded to the department.
A “due tomorrow” warning reaches the office before it breaches.
It is marked overdue, and the alert repeats until someone acts.
Past the grace period it escalates to the senior desk.
Proof
35+
AI agents running in our own production
Questions we are asked
That is what it is built for. Pages go to a vision model rather than classic OCR, which is why cursive Marathi on a Gram Panchayat letterhead works where a scanner-based system does not. Anything the model is unsure of is flagged for a person to correct, never guessed.
Only the roles you allow. Operators see the full record, viewers see contact details masked, and every view and edit is written to an audit trail against a named user. Nothing is handed to an outside vendor.
On infrastructure you choose — a server in the office, or a cloud region you name. The documents, the database and the credentials for WhatsApp and email all stay inside that boundary.
No. They send a photo of the letter on WhatsApp, or receive a tracking link by SMS. The link opens the status, the registration date and the expected resolution date in any phone browser — no app, no account, no password.
Yes, and most offices should. Scan-and-track is the smallest useful system: documents in, cases out, deadlines chased. WhatsApp intake, the citizen status link and the monthly insight brief can each be switched on later without moving the data.
Yes. Every case belongs to an office, so a constituency office, a municipal ward and a Zilla Parishad department can run in one deployment without seeing each other’s records. Categories, deadlines and alert recipients are set per office.
Where to go next
Documents
The extraction and verification pipeline behind the petition reader — fields lifted from a document, checked, and returned as structured data.
Reporting
The same digest habit pointed at the rest of the office — the monthly numbers assembled and written up before the review meeting.
Start here
Two weeks, one working agent on your own documents, and a costed plan for the rest — before you commit to a programme.
Free presentation · 13 slides · PDF
The presentation we walk boards and CXOs through — thirteen slides on what deploying AI agents actually involves, from the roles to the guardrails to the commercials.
No cost. Work email and organisation only.
Ten letters, scanned exactly as they are. You will see the fields, the categories and the deadlines the system would have set — before you commit to anything.
Or call +91 95525 86428
