1. Ingest
Open weather, satellite scenes, news, and department files enter through a queue. Paid feeds are added later as keys, not as a new architecture.
Platform
A base that can take the next department without a rewrite.
Pragna OS is the product. Modules are how a buyer meets it. Flood, crop, relief, and media forensics sit on the same routing, storage, and review path.
2. Route
The orchestrator reads the task, calls the relevant agents, and fuses their outputs into one structured result.
3. Verify
The Red Team checks required fields, numeric ranges, and unsupported language. A failed check is marked, not silently shown as fact.
4. Deliver
The decision-maker view shows the alert, the evidence, the confidence, and the recommended action. A human still signs off.
5. Audit
Task history and logs stay with the deployment. Department data is not hardcoded into a public frontend.
6. Extend
A new module is a new agent and a new data contract. The base stays. That is how traffic, pollution, or a state scheme can be added later.
Technology
The current build is a Python service layer with an agent registry, a task API, Redis for the queue, and a local analytical store. Cloud use is burst inference, object storage for imagery, a managed PostgreSQL database, containers for the orchestrator, and key management so credentials never sit in the browser.
We use open models for development and can place inference inside the buyer’s account when a department requires it.