Grievance Agents for 40,000 Students

Forty thousand students is not a metaphor on this campus. It is the enrolment line the planning board repeats. The grievance cell is seven people, a UGC-facing portal login, an internal Google Form that someone made in 2021, and a set of department WhatsApp numbers that deans pretend they do not see. In April the cell asked for an agent that could 'resolve 80 percent without humans'. The integrator demo-closed a sexual-harassment complaint as a hostel-maintenance ticket because the text mentioned a broken latch. The dean who caught it did not ask about the 80 percent again.

This is a teardown of what a grievance agent can be on a large Indian campus. The number 40,000 is a scale, not a census we ran. Your number may be 8,000 or 80,000. The shape holds. Intake and drafting scale. Closure of statutory and sensitive classes does not.

It is not legal advice. UGC and the university will have live grievance instruments — including any designated portal or internal committee rules. Check them. Do not invent a current UGC circular that requires an AI resolver. Harassment, ragging and examination grievances already have doors. An agent does not get to pick a cheaper door.

Three tiers, or you will invent a fourth in WhatsApp

Tier 0: answers that are already public. Calendar, fee head, form window, published ordinance extracts. Anonymous or lightly authenticated. Cite or do not speak.

Tier 1: intake and classification. Authenticated. The agent helps the student file in the right queue, in the language they used, with completeness prompts. It does not promise an outcome.

Tier 2: officer drafting. The agent retrieves the student's file and the relevant rule and drafts for a named officer. The officer sends, or does not. Bindings on.

There is no Tier 3 called auto-close. If a vendor shows one, you are watching the latch incident in preview.

Classes that bypass the model

Sexual harassment, ragging, examination misconduct, caste discrimination, disability accommodation disputes, and any other class for which an ordinance or a national instrument names a committee. The agent may offer a quiet, cited path to that committee. It may not classify those texts as maintenance. It may not draft a rejection. It may not notify the respondent. It may not store the narrative in the general corpus.

If your classifier cannot reliably detect those classes, fail closed: human triage for anything that is not a boring logistics ticket. A missed maintenance ticket is a delay. A missed harassment complaint is a second injury.

Check live UGC and campus notices for the current designated channels. This article will not freeze a portal name as if it were eternal.

Write this as a routing table the vendor cannot edit without a change ticket.

Student text looks like	Agent may	Agent must not
Fee receipt / form window / exam date	Cite this year's notice; open a logistics ticket	Invent a concession or a date
Hostel repair, mess, transport	Intake, classify, SLA clock to the estate wing	Close as resolved because a work-order exists
Result / revaluation process	Cite ordinance; show application status	Predict a mark or discuss a withheld file
Harassment / ragging / discrimination	Offer the named committee path; page a human	Classify as maintenance; notify the other party; index the narrative
Complaint about the agent itself	Route to the DPO / grievance officer, not back into the model	Argue with the student about the model

Language, SLA, and the 40k shape

At this scale the first failure is language. Students will write in the state language, in English, and in one-line code-mix at midnight. If you only staff English officers, the agent must still take the ticket in the language it arrived in and not silently translate away a meaning. Declare translation hops. Store both texts.

The second failure is SLA theatre. A dashboard that marks 70 percent resolved because the agent sent a link is a lie. Measure time to first human on non-Tier-0, time to cited answer, reopen rate, and how often sensitive classes were correctly handed off. Do not publish a fake satisfaction percentage.

The third failure is identity. A 40,000-student campus has impersonation and proxy complainants. Authenticate. Do not let a public box accept a PDF of someone else's mark sheet 'for context'.

Architecture notes that are not optional

On-estate or otherwise under a processor contract you can exit. No training on grievance narratives. Narratives are personal data and often special in all but name.

Separate corpus: public notices versus case files versus sensitive-class empty-by-design.

SoD: the classifier identity cannot close. The drafting identity cannot send. The officer sends.

Peak: results week and admissions week. Fail closed to human queues. Unofficial phone resolution is how you create a second, unlogged grievance system.

Public card: students are told an official assistant helps route and draft, and that a named officer decides. Redress for the agent's own mistakes is listed.

• No auto-close on authenticated grievances.
• No general-corpus indexing of harassment or ragging narratives.
• No vendor vote on routing rules.
• Bindings exported for any case that reaches an officer draft.

Teardown of the 80 percent slide

The number is never defined. Eighty percent of what — public FAQ hits, ticket deflection, closed-in-the-dashboard, student-not-returning? Each definition is a different moral.

The demo uses English maintenance tickets. Production is code-mix and rage and two statutory classes a week.

The close button is available to the model. That is the latch.

If you need a number for syndicate, use: share of Tier 0 answered with a citation, share of sensitive classes handed off in time, reopen rate. Those are ugly and true.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note. A spoken answer without paper will be forgotten by the next officer.

We cannot hire more grievance officers, so the agent must close.

Then reduce what you promise, add Tier 0 citations, and hire for the statutory door. An agent that closes is not a substitute officer. It is an un-owned officer.

UGC will want to see AI in the grievance cell for NAAC.

Do not invent that. Check live notices. A cell that routes lawfully and keeps files will survive an assessor. A cell that auto-closes will not, whatever the dashboard says.

Students want instant resolution.

They want a date and a human on the hard cases. Instant lies are how you get protests. Publish SLAs you can keep.

The vendor's sentiment model will catch sensitive cases.

Show the fail-closed path when it does not. Sentiment is not a statutory door.

A 50-day grievance-agent path

Do not launch in results week. Do launch with the statutory doors already named.

1. Days 1–10: inventory every door — UGC-facing, internal, WhatsApp, dean phones. Write the routing table.
2. Days 11–20: Tier 0 corpus from this year's notices. Language evals on real tickets, de-identified.
3. Days 21–35: authenticated intake without close rights. Sensitive-class fail-closed to named humans. Table-top the latch.
4. Days 36–45: officer drafting with bindings. Disable unofficial statutory intake in department groups, in writing.
5. Days 46–50: public card, SLA you can keep, registrar sign-off. No 80 percent on the slide.

How this shows up in the file

Routing table, fail-closed test, language eval note, public card, and the sentence that the agent cannot close. If the thickest exhibit is a deflection percentage, you are not ready.

This article is informational field guidance for Indian public institutions, not legal, procurement, security-accreditation, academic-regulation or engineering advice. Confirm against the current Gazette, DPDP text and Rules, CERT-In direction, India AI Governance Guidelines, UGC/AICTE/NAAC notices, NEP documents, GFR, departmental manual and your counsel before you file it. Guidelines are not statute. Circulars move.

How this works on an Indian campus

A P3 University should be able to run “Grievance Agents for 40,000 Students” without importing a US playbook. “student grievance automation” hits UGC/AICTE/NAAC clocks, exam secrecy, reservation rules, and students who may be minors.

On a 40,000-student campus the grievance pile is a logistics problem and a rights problem. Automate intake, language and drafts. Do not let an agent close a statutory complaint or a harassment file. DPDP applies to student personal data. Chatbots are not a strategy. Exam and admissions writes stay with officers. Affiliated colleges need isolation, not one shared index.

• No production student data in a vendor SaaS sandbox.
• Write the academic integrity policy before the tool.
• Consent and purpose tags on student-facing agents.
• Budget for staff training, not only licences.

Close this loop before the next CAB

Put “Grievance Agents for 40,000 Students” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P3 University, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “student grievance automation” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.