Field Operations

Field Operations & Labeling

Your capture teams use DataBridge Tag at the point of data collection — whether that is a robotics lab floor, a film set, a drone landing zone, or a field research station. When a storage device is filled, the operator launches Tag on their laptop or tablet. Tag detects the inserted device via USB, assigns a unique tracking label following your naming convention (e.g., HA-B047-SD-0193 — project code, batch number, media type, sequential ID), captures metadata (batch ID, operator identity, GPS coordinates from the host machine, free-text notes), and submits everything to the backend API.

Tag works fully offline. If the field site has no internet connectivity — common in remote research stations, underground facilities, or outdoor shoots — metadata is queued locally in an encrypted SQLite database and synced automatically when the device reaches a network. The label is the single identifier that ties every downstream action — buffer copy, cloud upload, integrity verification, archival — back to the original capture event. Nothing moves through the DataBridge pipeline without a label. This chain of custody is critical for teams working with egocentric capture data for AI training, where data provenance determines whether a dataset is usable for model training.

Tag supports batch workflows: an operator can label 50 SD cards in a session, assigning them all to the same batch (e.g., B047 — Day 12, Urban East Capture). The batch groups devices logically for downstream tracking, so when 50 cards arrive at the datacenter, operators and managers can see the entire batch progressing through ingest as a unit rather than tracking individual devices.

Datacenter Ingest

At the datacenter, operators plug labeled devices into the ingest server's USB hub. DataBridge Core detects the new device within 2 seconds via udev event monitoring, validates the tracking label against the backend API, and begins the ingest workflow automatically. The workflow has four stages: buffer copy to NVMe (sequential read from USB at up to 10 Gbps per port), SHA-256 manifest generation (every file hashed, manifest written to buffer alongside the data), queued upload to cloud (S3 multipart upload with configurable part sizes, typically 64-128 MB for optimal throughput), and post-upload verification (ETag comparison against the local manifest).

Each server runs 5 concurrent ingest lanes. While Lane 1 uploads a 2 TB HDD to S3 over the Direct Line at 9+ Gbps, Lanes 2-3 can copy SD cards from USB to the NVMe buffer, Lane 4 can run post-upload ETag verification on a completed transfer, and Lane 5 waits for the next device in the queue. The operator monitors everything through DataBridge Ops on their workstation — real-time lane status, queue depth, throughput per lane, buffer drive utilization, and estimated time to completion. If a device needs to be prioritized (e.g., urgent footage for a client deadline), the operator drags it to the top of the queue in Ops.

For high-volume operations processing hundreds of devices per day across 10+ servers, the fleet-wide view in DataBridge Watch shows aggregate throughput, per-server health, and alert status. The entire ingest process — from device plug-in to verified cloud archive — is fully automated. Operators are there to manage the physical workflow (plugging and unplugging devices) and handle exceptions, not to babysit software.