Affiliated Colleges: Multi-Tenant AI Design

An affiliating university in a large state may have hundreds of colleges, some government, some aided, some private, some with their own NAAC grade and their own AICTE EoA. The computer centre, under deadline, stood up one agent with one retrieval index: 'all circulars and all student stores'. A principal in College A asked a harmless question about a roll number that also existed in College B. The agent, being helpful, answered from College B. The two students were not the same person. The parents were.

This guide is for registrars of affiliating universities, college principals, and the SI who will otherwise sell you a single brain. Multi-tenant design is not a cloud buzzword here. It is the difference between a platform and a leak. The university may be fiduciary for exams and degrees. The college may be fiduciary for attendance, hostels and local fees. Sometimes they are joint. None of that is permission to mix the indexes.

Not an affiliation-regulation treatise. UGC and state Acts describe affiliation. NAAC assesses colleges and universities on different manuals. AICTE approves technical programmes at the institution it names. Your agent should know those boundaries as tenant walls, not as trivia.

Who is fiduciary for what

Write a matrix before any shared agent. Typical split: university owns tabulation, degree, central admission where it exists, and university circulars. College owns local attendance, college fees, hostel if any, internal exams that are not the university exam, and college IQAC evidence. Joint: some admission counselling, some scholarship verification, some affiliation returns.

A shared agent for a joint purpose still needs a tenant key on every row. Joint is not 'everyone sees everything'. Joint is 'this workflow may see these fields for these students of this college'.

Minority colleges, autonomous colleges, and constituent colleges are not the same tenant type. Autonomy in particular means the college's academic store is not yours to index 'for convenience'.

If you cannot name the tenant on the row, the row does not go in the index.

Workflow	Default tenant	University may see	Other college may see
University exam / revaluation	University, keyed by college code	Yes, for that student	No
College internal marks	College	Only if ordinance says so for moderation	No
College hostel tickets	College	No	No
Affiliation / AICTE pack	College, with university view of declared fields	Declared fields only	No
NAAC evidence	The assessed entity	Not another college's raw personal data	No

Isolation is engineering, not a prompt

Tenant ID on every document, every embedding, every log. Separate indexes or a mandatory filter that cannot be overridden by a prompt. Prompt-level 'do not reveal other colleges' is not a control. It is a hope.

Identity: college staff authenticate to their tenant. University exam staff authenticate to the exam tenant. A principal does not get a university-wide search because they sit on a board. Break-glass access is logged and time-bound, like production access anywhere else.

Models may be shared. Indexes must not be. A shared on-prem model with a training ban and per-tenant retrieval is the shape. A shared chat history is how College A's ragging complaint appears in College B's 'similar cases'.

Commercial and political pressure to merge

Vendors like one index because it is cheaper. Ministers like one dashboard because it is a slide. Principals like to see 'peer' numbers. Peer numbers, if you must, are aggregates the college already reports. They are not another college's student rows.

A state directorate may demand a roll-up. Give a roll-up table the university already knows how to produce. Do not give the directorate an agent that can drill to a mobile number across tenants unless a law says so and you have written it.

When a college disaffiliates, the tenant leaves. Exit is not optional. Their local store goes with them under the agreement. The university keeps what the Act says it must keep — usually the official result. Design that split on day one.

NAAC and AICTE are per institution

Do not build a 'quality agent' that hoovers every affiliated college to write the university SSR. The university SSR uses university evidence and the affiliation data it is entitled to. College SSRs stay in college tenants.

AICTE EoA is filed by the approved institution. The university agent may help a college assemble a pack inside the college tenant. It may not 'borrow' faculty from another college to fix a ratio. That failure mode is documented in the AICTE piece in this cluster. Multi-tenant walls are how you prevent it at scale.

Identity issuers and roll formats

Universities issue enrolment numbers. Colleges issue class rolls. Entrance agencies issue application IDs. If the agent searches any ID without a tenant and an issuer, it will collide. Write the issuer on the key. Display the issuer to the clerk. 'Roll 4418' is not an identifier in an affiliating system. It is a coincidence waiting for a parent.

Staff who teach in two colleges are two identities, or one identity with two tenant memberships that do not share student stores. Do not merge them because the PAN is the same. Merging staff is how a principal in College A opens College B's internals 'by accident'.

Logs must carry tenant_id or they are not logs you can show a court. A prompt-injection that crossed a wall should be reconstructable. If the log store is shared without a tenant field, you will not know what leaked. That is the CERT-In conversation and the syndicate conversation in one sentence.

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.

One model is more accurate if it sees all data.

Accuracy at leaking the wrong student is not a metric we want. Per-tenant retrieval is the constraint. Shared weights without shared indexes are enough.

The university is one legal person.

Sometimes. Colleges are often other legal persons. Even when they are not, purpose and need-to-know still split the stores. A single legal person can still isolate.

Small colleges cannot run tenants.

The university can host the platform and still isolate. Hosting is not mixing. Managed isolation is the product.

NAAC peer teams want comparative data.

Give comparative indicators the colleges already publish. Do not give peer teams another college's personal files through your agent.

A tenant design you can draw on one board

If you cannot draw it, you cannot buy it. Markers first, vendors second.

1. Week 1: fiduciary matrix and tenant types (constituent, affiliated, autonomous, minority).
2. Week 2: identity, index, log and break-glass design. Ban global search.
3. Week 3: pilot two colleges plus the exam tenant. Try to leak. Fail on purpose.
4. Week 4: disaffiliation exit rehearsal. If exit fails, the design fails.

How this shows up in the file

Affiliation-section note: tenants, keys, no global search, shared model allowed, shared index forbidden, exit on disaffiliation, NAAC/AICTE packs stay in the assessed institution's tenant.

Attach the failed leak test. A passed test you did not try is not a test.

This article is informational field guidance for Indian public universities and colleges, not legal, regulatory, academic, labour or procurement advice. Confirm against the live AICTE / UGC / NAAC notice, the DPDP Act and Rules as commenced, your Act and statutes, reservation orders, labour counsel and the syndicate before you file it.

How this works on an Indian campus

A P3 University should be able to run “Affiliated Colleges: Multi-Tenant AI Design” without importing a US playbook. “affiliated college system AI” hits UGC/AICTE/NAAC clocks, exam secrecy, reservation rules, and students who may be minors.

A shared vice-chancellor is not a shared student record. Design tenants first. The model is the last thing you share. 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 “Affiliated Colleges: Multi-Tenant AI Design” 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. “affiliated college system AI” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

What the next file must contain

“Affiliated Colleges: Multi-Tenant AI Design” earns a line in the noting only if a P3 University can attach proof of “affiliated college system AI.” A heading is not proof. A vendor slide is not proof. A workshop photograph is not proof.

Write three dated sentences: what was decided, who owns it after the next posting order, and when it will be re-checked. If you cannot write the three sentences, you are not ready to buy, to sell, or to go live.

Leave unsourced percentages out of the note. DPDP is not a blanket localisation statute. The November 2025 AI governance text is guidance, not an Act. CERT-In’s 28 April 2022 directions still set specified incident and log clocks. A PAC, when lawful, lives in GFR Rule 166.

• Name the designation that owns “affiliated college system AI.”
• Attach one artefact a stranger can open next year.
• Record the instrument you are actually using.
• Revisit when the model, the SI, the notice or the posting changes.