| Tender ask | Ganga's answer, as shown in this sample pilot | Fit |
|---|---|---|
| AI compute for government workloads | On-premise AI inference on the facility's own attested nodes. Personal data never leaves the building; every job runs behind a DPDP consent gate and produces a hardware-signed compliance certificate. | Demonstrated |
| Compliance & audit for a government auditor | Every certificate is verifiable against a hash-chained, checkpoint-verified audit ledger. Audit packs per customer: jobs, certificates, consent blocks, incidents, grievances - one pull. | Demonstrated |
| Operations & resilience (O&M) | Nodes that go offline queue their jobs and drain on return - no lost work. Key rotation with full history. CERT-In incident reports inside the 6-hour window, linked to affected certificates. Grievance redressal against a published 72-hour SLA. | Demonstrated |
| Disaster recovery discipline (backup, replication, restore, RPO/RTO) | Point-in-time, integrity-hashed snapshots of a customer's recoverable state replicate to a second attested node - a replica that cannot reproduce the state hash is rejected. Restore drills bring verifiable service back from a replica and measure RPO and RTO; a drill that cannot re-verify every certificate FAILs. Every step is a hash-chained ledger entry. Pilot to date: 876 snapshots, 876 verified replicas, 876 restore drills passed (last RPO 0.02s, RTO 0.0s). Civil and power redundancy stay with the consortium's infrastructure partners. | Demonstrated |
| Civil works, power, racks, GPU hardware, training | Not Ganga's layer. Ganges.ai joins as the sovereign-AI + compliance technology partner inside the consortium; civil, power and hardware partners cover the physical build. | Consortium |
Every compute node in the facility proves its identity with its own hardware key before it accepts work. A node that cannot attest does not run - that is the sovereign boundary.
The job is checked against a registered DPDP consent artefact: purpose, data class, scope, validity window. It runs on-premise; personal data is destroyed with proof; the node and the Agni-Shield co-sign a compliance certificate. The auditor's paper is a by-product of running, not an afterthought.
A node drops offline. Its jobs queue, visible in telemetry with queue age, and drain automatically when it returns - in the latest 2,000-user pilot wave: 691 outages, 691 clean drains, zero lost jobs.
A point-in-time, integrity-hashed snapshot of the customer's state is replicated to a second attested node, which must reproduce the state hash before it counts as a replica. A restore drill then brings service back from that replica and re-verifies its certificates, measuring RPO and RTO on the record: 876 drills passed so far, last RPO 0.02s, RTO 0.0s. A drill that cannot re-verify every certificate FAILs - and that failure would sit on the same hash-chained ledger as everything else.
An incident is declared. The CERT-In report artefact is generated inside the 6-hour statutory window and linked to the affected certificates, which are revoked - verifiable by anyone, instantly.
A data principal files a grievance against a certificate. Resolution is tracked against a published 72-hour SLA, and breaches are flagged. Redressal is an operational system, not a mailbox.
One call produces the audit pack: jobs, certificates, consent blocks, incidents, grievances. Every certificate verifies against the hash-chained ledger - valid, revoked or expired, with the reason.
A healthy facility job certificate:
{
"verdict": "VALID",
"certificate_id": "04f2d5fd-98f8-48a5-a35a-c1083bf5f644",
"chain_head_seq": 19195,
"latency_ms": 27.5
}
After a consent-withdrawal cascade:
{
"verdict": "REVOKED",
"certificate_id": "9b905cbf-b1e6-48fd-a31a-b424604fc18b",
"reason": "post-hoc incident drill",
"latency_ms": 0.7
}
Ganges.ai supplies the sovereign-AI and compliance layer for the consortium bid: the architecture description (attested nodes, Agni-Shield co-signing, consent-gated jobs), the compliance framework (DPDP certificates, CERT-In reporting, grievance redressal, disaster-recovery drills with measured RPO/RTO, audit packs), and the node hardware specification. Delivered same-week on request; a live demonstration runs from this pilot.