Low Voltage · Data Center
Trained the symbol library to recognize their proprietary structured-cabling patterns.
A specialized low-voltage shop builds structured cabling for data centers. Their drawings use proprietary symbol conventions — generic AI tools struggled to recognize their fixtures correctly out of the box. The active-learning loop trained a custom recognizer over the first weeks of real bid work.
Industry
Data Center
Region
Texas
Annual revenue
Mid-market
Headline numbers
- Recognition accuracy (baseline)
- Customer-validated outcome
- Recognition accuracy (post active-learning)
- Substantial improvement
- Training annotations
- Multiple reference projects
Generic vision-only starting point
Specific accuracy curves shared under NDA
Annotations seeded the recognizer
Challenge
What they were solving for
Data-center structured-cabling drawings use highly-specialized symbols that are not in any public training set. Generic AI was misclassifying enough fixtures that they couldn't trust the output.
Solution
How OmniTakeoff worked
Active-learning loop: vision claim → review queue → confirm/reject → ground-truth annotation → re-train. The loop was fed across multiple reference projects over the first weeks of bid work.
Outcome
What changed
Recognition accuracy improved meaningfully on their drawing style and the model keeps learning on every confirm. Specific accuracy curves and bid-cycle compression are shared under NDA on reference calls rather than published.
We feed it our drawings, we click the wrong calls, and within a few weeks it just gets our symbols. Being able to teach the model on our own dialect is the thing that mattered.
Your numbers next
Apply to pilot.
Four-week pilots on real bids. We commit to specific outcome targets up front.