Model Registry
Manar maintains registered models across six families. Every one of them started as an open foundation model and was trained further on a real site's own conditions.
Trained vision over cameras, imaging, and physical sites.
Reads vibration, pressure, grid load, and building telemetry as it streams.
Documents, curricula, and farm records, retrieved and reasoned over in Urdu or English.
Climate risk and environmental monitoring read across a property's own ground.
Acting models for supply chain and commerce platforms, scoped to one operation.
The same method, trained on whatever the client's operation actually is.
Reads informal Pakistani text, Roman Urdu, mixed-script, colloquial, and extracts the meaning before any document is searched or action is taken.
Every code follows the same order: family, then domain, then method. SIMAH-PERIMETER-VT is a Simah vision model, trained for perimeter and site monitoring, using a vision transformer method. The structure is fixed so any model in the registry can be identified from its name alone, without needing to ask what it does.