Energy / Pipeline NDT · Fathom Seam
91.2% per-pore recall, where frontier vision models score 0-4.1%
The situation
WhiteWater Midstream inspects welds across its pipeline network, working with engineering partner LSC. Weld film review has always been a bottleneck: a human has to look at every foot of film, and a miss is expensive. WhiteWater signed a Fathom Seam pilot to test whether an AI-plus-physics-plus-ruleset pipeline could catch what people miss, without adding cost per weld and without ever approving a bad weld on its own.
What Fathom did
- Deployed Fathom Seam's three-stage pipeline: AI classification, physics measurement, deterministic API 1104 ruleset decision
- Fails closed to a human auditor on anything the ruleset can't resolve
- Runs fully offline with no per-weld token cost
- Outputs risk-tiered results (crack-like, volumetric, benign) plus welder-performance analytics
- Ingests film automatically through a hot folder
Inside the product
AI Classification + Physics Measurement
Two independent passes over every film before a decision is made.
API 1104 Ruleset Engine
A deterministic ruleset decides, not a model guess.
Human Auditor Fail-Closed Review
Anything the ruleset cannot resolve routes to a person, every time.
Hot-Folder Ingestion
Film drops in, results come out, no manual upload step.
Welder Performance Analytics
Risk-tiered output rolled up by welder over time.
Timeline
See what Fathom could do for your team
More customers