Physical environments

Start with the observation

A workload can begin with a sensor record, an image, a video frame or structured operational data. Keep record identifiers stable and attach the files each worker requires. Separate observation from action so outputs can be inspected before another system consumes them.

Typical environments include field systems, industrial equipment, ports and distributed sensing. The same job interface applies even when the underlying hardware changes.

Run a bounded image workload

A predict job applies a served detector to each record’s image without a custom worker. This example submits one harbor tile, keeps only the ship class, and retrieves the detection rows.

1
2
3
4
5
6
7
8
9
10
11
12
13import drkm

drkm.connect()
records = [
    {"id": "harbor-01", "files": ["tiles/harbor-01.png"]},
]

job = (drkm.parallelize(records)
       .predict("overwatch-obb-r3-int8", classes=["ship"])
       .submit(name="Vessels in harbor imagery"))

job.wait()
print(job.rows())

Return useful results

Keep the result contract specific: detections, classifications, summaries, decisions or instructions for an authorized downstream system. Use record IDs to associate each response with its original observation.

Continuous work has a different lifecycle from a finite job. A Watchdog guard runs until you stop it: closing the browser does not end it, and an outpost that restarts or loses power resumes it under the same job ID. Design the receiving system around explicit state, coverage and freshness.

Continue with jobs and workloads for lifecycle details, or connectors to connect an operational system.