PSU & CPSE
Multi-Location PSU Rollouts: Coordination
· 10 minute read
Thirty plants will not become one Kubernetes mesh. Ship a versioned appliance, a media bus, and the humility to let a weak site stay on a smaller pattern.
Corporate IT promised one agent for nineteen locations. The BoQ was nineteen identical GPU mini-clusters. Three sites had a hall and a cooling story. Nine had a comms room and a diesel story. Seven had a desktop and a clerk who still posted paper. The first review after Diwali showed three live sites, a mutiny in the west, and a corporate dashboard that counted 'rollout percent' by hardware received, not by a working packet.
This playbook is for the SI and the corporate programme office that have to put the same purpose on many PSU sites without pretending the sites are equal. It is a cousin of the air-gapped multi-site note. Here the emphasis is coordination: who versions, who trains the local champion, who pays power, who talks to the local union, and who is allowed to say this site stays on the thin pattern.
A maharatna is a federation that pretends to be a hierarchy. Plant heads can stall a corporate banner without writing a letter. Your RACI has to include them as actors, not as recipients of a push notification.
Not a staffing plan and not a network design for classified networks. Draw the actual WAN. Write the actual owners.
Three patterns, not one mesh
Pattern A — local appliance: the site can hold a small box, has a local champion, and has a data class that should not ride the WAN. Pattern B — regional hub: several nearby sites point at a hub over the corporate WAN, with a documented RTO if the WAN dies. Pattern C — thin client: the site uses a browser to a corporate or SDC instance and keeps a paper fallback. Write which sites are which before you buy nineteen of anything.
Do not stitch A, B and C into a single Kubernetes fabric with thirty VPNs and a prayer. That fabric will become the programme. The purpose — retrieval of SOPs, invoice exceptions, CSR packs — will wait.
| Pattern | When it fits | Coordination load |
|---|---|---|
| A local appliance | Weak WAN, sensitive corpus, a hall that can take a box | Media bus, local champion, physical access |
| B regional hub | Decent WAN, shared language and corpus, a willing hub plant | Hub owner, chargeback, weekend RTO drill |
| C thin to centre | Clerk, desktop, public or low-sensitivity corpus | Identity, paper fallback, no fake 'edge GPU' |
Version one purpose, not nineteen experiments
Corporate owns the purpose, the prompt, the evaluation set, and the packet schema. Sites own access lists and the local corpus slice. If every plant fine-tunes a private dialect in month one, you will not be able to answer CAG or the CVO with a single sentence about what the agent is.
Ship versions on a calendar. A media bus or a hub pull, hashed. A site that skips a version is marked amber, not 'innovating'. Local prompt hacks are change tickets, not folklore.
The RACI plants will actually answer
Corporate programme: purpose, version, vendor contract, measurement. Plant head: pattern choice, local champion, union sentence, power and room. Local IT: identity, backup, the deny-outbound proof. Local user lead: corpus in-force list. AMS/SAP SI: any ERP connector. Vendor: the appliance and the packet export. If a name is missing, that row will become a WhatsApp group and then a stall.
Chargeback kills silent resentment. If plant B is the hub, plant B's cost centre needs a line, or plant B will deprioritise the box when their own closing week arrives.
Unions, languages, and the dignity of the weak site
A corporate Hindi-English corpus will fail a plant that works in Odia or Tamil on the shop floor. That is not a reason to promise 22 scheduled languages. It is a reason to add the one or two working languages that site actually uses, with an eval set, or to keep that site on Pattern C until you can. See the Indic opinion in this cluster's neighbour.
Local unions need the retrieval-not-replacement sentence in the local language of the notice board. A corporate email in English is not a briefing.
Let weak sites stay smaller. Forcing a GPU into a room that cannot cool it is how you get a dead asset and a plant head who will not take your next call.
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.
Corporate standard is one architecture.
One purpose and one packet schema is a standard. One SKU for every hall is a superstition. Write the exception path.
The WAN is MPLS, so everyone is Pattern B.
MPLS dies. Nightly batches eat the pipe. Measure, then choose. A labelled Pattern C is better than a pretend hub.
Local champions will create shadow IT.
They will if you give them no version. A named champion on a versioned appliance is how you avoid a second vendor in a drawer.
We will train every site the same week.
You will train three, exhaust the vendor, and create folklore at the rest. Wave the sites. Measure each wave's citation quality before the next crate moves.
A four-week coordination playbook (then waves)
Week 0 is honesty. Waves come after.
- Week 1: score every site A/B/C on hall, WAN, champion, union heat, and data class. Publish the map. Expect arguments. Better now.
- Week 2: write the RACI and the hub MoU. Name chargebacks. Kill the identical-BoQ if the map does not support it.
- Week 3: freeze purpose, packet schema, and version 1 on one Pattern A and one Pattern C site.
- Week 4: tabletop a hub outage and a union notice. Only then issue the first wave's movement order.
How this shows up in the file
Subject: Multi-site agent rollout — patterns, RACI, versioned purpose.
Sites are classified A/B/C (annexure). Corporate owns purpose, version and packet schema. Plant heads own pattern fit, champion and local briefing. Hub sites have a written MoU and chargeback. Weak sites may remain on the thin pattern without penalty. Hardware follows pattern, not vanity.
This note is not a network design. It is a coordination order.
This article is informational field guidance for Indian public sector undertakings and their vendors, not legal, audit, labour, energy-regulatory, banking-regulatory or procurement advice. Confirm the live circular, DPE guideline, CVC instruction, sector regulator text, purchase manual and your counsel before you file it.
How this clears vigilance and the board
A P4 System Integrator in a PSU will meet CVC-shaped questions even when there is no special 'AI circular'. “Multi-Location PSU Rollouts: Coordination” has to survive a technical committee, a cost centre, and a union conversation if jobs appear threatened.
Thirty plants will not become one Kubernetes mesh. Ship a versioned appliance, a media bus, and the humility to let a weak site stay on a smaller pattern. OT networks stay off-limits. Navratna autonomy speeds buying; it does not waive DPDP or data classification. IREPS is not GeM. RBI-shaped rules still localise payment data.
- Classify data before the POC.
- Keep agents off OT.
- Write the board memo with residual risk.
- Engage unions on retrieval vs replacement.
Close this loop before the next CAB
Put “Multi-Location PSU Rollouts: Coordination” 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 P4 System Integrator, not “the vendor.”
Revisit the item when the model, the GeM term, the region, or the SI changes. “multi location rollout PSU” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.
What must be true before you file this
If “Multi-Location PSU Rollouts: Coordination” is only a heading, it will not survive a file inspection. A P4 System Integrator should be able to attach one artefact that proves “multi location rollout PSU”: a log export, a clause, a scored row, a dated notice, or a refusal rule.
Write three dated sentences: what was decided, who owns it, and when it will be re-checked. Unsigned sentences are souvenirs. Dated sentences are controls.
- Name the owner of “multi location rollout PSU” inside the institution.
- Attach one artefact a stranger can open next year.
- Revisit when the model, the notice, or the SI changes.
- Do not treat a vendor slide as evidence.
What the next file must contain
“Multi-Location PSU Rollouts: Coordination” earns a line in the noting only if a P4 System Integrator can attach proof of “multi location rollout PSU.” 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 “multi location rollout PSU.”
- 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.
Questions this usually raises
- Should every PSU plant get the same AI stack?
- No. Same purpose and packet if the workflow is the same. Hardware and topology follow the hall, the WAN and the data class.
- Who pays for a hub site?
- Write a chargeback or a corporate line. Unpaid hubs go dark on closing week.
- How do we version air-gapped plants?
- A hashed media bus on a calendar, plus an emergency path. Daily USB folklore is how you lose the gap. See the air-gap companion notes.
- What if a plant refuses?
- Leave them on paper or Pattern C. A forced crate is a dead crate. Record the refusal so corporate dashboards stay honest.
- Do we need 22 languages for a national rollout?
- No. Need the working languages of the sites in the wave, with eval. Do not print 22 on the banner. Read the Indic opinion.