Provenance is not optional
Claims are immutable and each one points at a document, a source and a credibility score. Nothing enters project knowledge unattributed.
Construction Intelligence OS
Operator sign inInternal platform · India pilot
A construction intelligence operating system, built provenance-first.
Continuous discovery over public registries, government open data and news signal; deterministic filtering and extraction; immutable claims with conflicts kept in the open; and historical snapshots you can query as of any date. Designed for India first and for 190+ countries after, without forking the engine.
The pipeline
Each stage is inspectable in the console, with its own operational surface and failure modes.
- 01
Discover
Provider sweeps and query planning across registries, not a crawler.
- 02
Observe
Every fetch archived raw with hash, provider and timestamp.
- 03
Filter
Deterministic rules and hashes before any model is invoked.
- 04
Detect signal
Scored candidates, never unfiltered noise.
- 05
Extract
Claims and events with a citation for each assertion.
- 06
Resolve
Entity matching with human adjudication where ambiguous.
- 07
Score
Credibility-weighted confidence, conflicts kept visible.
- 08
Snapshot
Immutable historical state you can query at any date.
Design commitments
Geography and ontology are data
India geography follows Government of India LGD codes. Sectors, project types, stages and event types are registry rows — adding a country or sector needs no code change.
Deterministic before probabilistic
SQL, rules, regex, hashes and scoring do the volume work. Models are reserved for ambiguous classification, genuinely unstructured extraction and hard resolution cases.
Operations are first-class
Jobs, checkpoints, provider health and dead letters are visible surfaces, because a continuous discovery system is judged on what it does while nobody is watching.