How Our Models Are Built — and Who Owns Them

Built In-House, Before We Ever See Your Data

Every FloLink model starts as a base model developed entirely in-house, using our own proprietary training algorithms and internally sourced or synthetically generated imagery. These base models are already heavily fine-tuned to their target domain — leak detection, PPE compliance, gauge reading, and more — before a single client image is ever involved.

Our internal algorithms are what let us produce base models that are performant, lightweight, and purpose-built for a narrow, well-defined task — models small enough to run in real time on modest on-chip camera hardware, without a round trip to the cloud, and without requiring a data center's worth of GPUs to train.

What Happens Once a Model Reaches Your Site

A performant base model is only the starting point. For most deployments, we fine-tune that base model further using data captured directly at the client's facility, so it adapts to that site's specific cameras, lighting, equipment, and operating conditions.

This is the point at which our data ownership policy becomes absolute:

• The instant a base model is exposed to a single proprietary image from a client's site, that model is considered to have entered the client's premises.
• From that point on, the fine-tuned model — and any data used to tune it — never leave the client's premises and are never reused on another site or another client.
• That fine-tuned model is the exclusive property of the client. FloLink will not copy, redeploy, or share it elsewhere without the client's explicit permission.
• There is no shared training pool and no cross-client model. Each client's fine-tuned model exists in isolation, tied to that client alone.

Why This Matters