One fabric. Nine modules. Every domain.
Each module has a narrow job and a Sanskrit name. Together they take raw, fragmented source data to an accountable operational decision, without collapsing everything into one unrestricted database.
Security & policy, cross-cutting
Evidence & forensic audit, cross-cutting
Local runtime & sync, cross-cutting
Source to operational picture
Every material transformation between a raw source and a displayed object is versioned, and the original payload is never overwritten.
Source
Connector authentication and payload integrity validation.
Validate & classify
Malware inspection, classification and ownership tagging.
Preserve evidence
Raw payload stored immutably before any transformation.
Map to ontology
Canonical event creation and TATTVA ontology mapping.
Resolve entities
Candidate matches scored; ambiguous cases go to an analyst.
Display
Policy-filtered projection rendered in DRISHTI, fully audited.
SETU
Connectors, ingestion and provenance. Existing systems come in without forced replacement.
- Connector Manager and Source Adapter Runtime
- Schema validation and classification tagging
- Malware quarantine and data-quality scoring
- Full lineage from raw source to derived object
TATTVA
A shared operational language for units, assets, missions and events, extended per Service, never overwritten.
- Entity, relationship and entity-resolution services
- Temporal and geospatial reasoning
- Versioned schemas, rules and relationships
- Object-, field- and relationship-level access control
DRISHTI
One geospatial view of every authorised entity, with freshness, confidence and classification always on screen.
- Map tiles, layers and timeline replay
- Low-bandwidth and fully offline modes
- Stale or contradictory data highlighted automatically
- Export and briefing-package generation
MEDHA
An indigenous model core plus an approved catalogue, spanning vision, forecasting, anomaly detection and graph analytics.
- Self-hosted, indigenously trained model core, not a foreign API
- Thousands of domain-scoped RAG swarms instead of one generic model
- Computer vision for OCR and change detection across imagery
- Circuit breakers on drift, latency and uncalibrated confidence
SANJAYA
Answers only from authorised, retrieved evidence, with citations and confidence attached, never an invented fact.
- Swarm retrieval across every connected source, ranked and reconciled
- Citation and structured-output validation on every answer
- Prompt-injection filtering and strict tool allowlists
- Human confirmation before any state-changing action
VYUHA
Tasks, incidents, approvals and escalations. No AI recommendation can move a workflow into an irreversible state alone.
- Configurable workflow states and role assignment
- Two-person approval for high-risk actions
- Offline execution with later reconciliation
- Escalation and after-action review services
KAVACH
Zero-trust identity and policy enforced at every layer, with classification rules that travel with the data.
- Identity, workload identity and privileged access
- Policy decision and enforcement points
- Device trust and classification enforcement
- Cross-domain guard integration
SMRITI
Every model call and consequential action is written to a tamper-evident, hash-chained ledger investigators can replay.
- Append-only, hash-chained ledger
- Evidence manifest and chain-of-custody export
- Authorised replay of significant decisions
- Retention and archive management
EDGE
Local map, workflow and inference, running at least seven days disconnected, then reconciling on recovery.
- Local policy engine and inference runtime
- Encrypted evidence capture and store-and-forward
- Signed update import with rollback
- Health and attestation monitoring
Not one model. A swarm of them.
MEDHA does not route every question to one generic language model. It runs thousands of small, domain-scoped retrieval swarms in parallel, one grounded in maintenance logs, another in coastal AIS, another in signals traffic, on an indigenous model core self-hosted inside sovereign infrastructure.
Thousands of RAG swarms
No single generic model answers every question. Thousands of small, domain-scoped retrieval swarms work in parallel, one grounded in maintenance logs, another in signals traffic, another in coastal AIS, then reconciled and ranked before an answer ever reaches a screen.
An indigenous model core
MEDHA is built around VISHVAKSHA's own trained models, self-hosted inside sovereign infrastructure alongside an approved open catalogue. No mission-critical inference leaves through a foreign API.
Computer vision & OCR
Scanned reports, handwritten logs, satellite frames and document backlogs become structured, searchable evidence, correlated into the same ontology as every other source.
Sensor & surveillance fusion
Telemetry, video feeds and field sensors from every connected platform stream into one picture, change-detected and anomaly-scored before an analyst ever has to look.
Illustrative: parallel retrieval swarms, ranked and reconciled before synthesis.
See it mapped to your data.
Bring your source systems. We'll walk through the connector and ontology fit.
Request Briefing