Urvarā · Digital Twin
The schema, in and out
REAL DATA · open geospatial

BIMSaarthi Technologies · Amaravati · Andhra Pradesh

Urvarā

Unified Real-time Vision & Action for Resilient Agriculture

A semantic digital twin of Indian agriculture — every parcel, season, sensor and scheme rendered as meaning, not just data: an OWL ontology, gated by SHACL, reasoned by SPARQL, and audited end-to-end by PROV-O. This is the schema, in and out.

⬡ Open the interactive KG Explorer →

scroll · the thread unwinds ↓
01 / THE SUBSTRATEsūtra — the thread that binds knowledge

Sutra — the semantic engine Urvarā is woven on

Urvarā is not a database with a schema bolted on. It is built on Sutra — a reusable semantic substrate of five interlocking threads. Each thread does one job the others can't; together they make every fact meaningful, validated, derivable, and auditable. Urvarā is the agriculture instantiation; the substrate is domain-agnostic.

Urvarā
42 agricultural domain ontologies — crop · field · soil · water · weather · health · insurance · market · governance · … — woven on the five Sutra threads
02 / THE ATOMinvariant #10 — every query starts here

The atomic unit: FieldParcel × CropSeason

Every observation, inference, advisory and claim in Urvarā hangs off one cell: a parcel in a season. A satellite reading observes this parcel, this Kharif; a drought index is computed for this mandal, this window; an advisory is rendered for this cell. The schema makes the cell the spine — so a single SPARQL pattern reaches the whole twin.

From the cell radiate the four observation families and their derived conclusions — each a typed class, each carrying provenance back to its source.

03 / THE ONTOLOGY42 modules · 235 classes · 293 relations · 1,262 attributes

A modular ontology — the whole farm, formalised

The vocabulary is split into 42 OWL 2 DL modules, one per domain, importing a shared core. Reasoned by HermiT, version-controlled with ROBOT. Hover any module.

04 / IN & OUTdeclared vocabulary ↔ living instances

The schema is alive — populated, not just declared

A schema is only as real as what fills it. These are the most-populated classes in the live graph — sized by instance count. The twin breathes: 67k climate readings, 14k drought indices, 1,710 crops fully described, 950 real parcels. Hover a class.

05 / THE LAYERS24 named graphs · ~1.36M triples · conclusions, not computation

Named graphs — the digital twin in strata

Data is layered into 24 named graphs, each a stratum of the twin. High-rate raw streams (IMD weather, NDRE) live as Parquet off-graph; only audit-grade conclusions become triples — each prov:wasDerivedFrom its source. Bars are log-scaled by triple count.

06 / INTEGRATIONpixel → gate → graph → rule → agent → advisory

How the world becomes knowledge

Heterogeneous sources are pulled by orchestrated DAGs, forced through the SHACL gate, layered into a named graph, reasoned by deterministic SPARQL, and turned into a farmer advisory by a stateful agent — de-identified, in Telugu. Determinism is the legal differentiator.

07 / THE SPINEthe three non-negotiables

Gate · Rule · Provenance

Three invariants make Urvarā court-defensible: nothing enters un-validated, every decision is deterministic, every fact traces to its pixel.

08 / INTEROPa layer over the national agri-stack, not a replacement

Plugged into India's machinery

Urvarā interoperates — it makes existing government systems auditable rather than replacing them. It feeds the PMFBY 30% technology weightage, reconciles YES-TECH against satellite evidence, and routes payouts only through the IRDAI insurer.