Qcvisionly started inside Ann Arbor’s automotive tier-1 corridor, where our founders spent years watching the same failure cycle repeat: a surface defect escapes manual inspection at line speed, a SCAR event arrives from the OEM, and the corrective action adds another inspector to a station that already has three on each shift.
Priscilla Nakamura spent three years as a quality systems engineer at a Tier-1 stamped metal supplier in the Ann Arbor corridor. During a 14-week stretch, the line generated four consecutive SCAR events from the same OEM customer—all traced back to surface defects that post-analysis confirmed were present on parts that had passed manual inspection at rated line speed. Priscilla personally reviewed every escape report. The defects were not subtle failures of inspector attention. They were 0.2 mm surface scratches on parts moving at 35 units per minute under overhead fluorescent lighting.
That’s when the insight landed: the inspectors were not failing at their jobs. They were being asked to perform a task that human visual physiology cannot sustain at production line speed. At 35 units per minute, an inspector has under 80 milliseconds per part. Fine surface defects at that resolution and cadence are simply beyond what the eye can reliably detect across an 8-hour shift, let alone three shifts. The inspection failure was a system design problem. Adding a fourth inspector would not change the physics.
The first attempt was deliberately low-budget. Priscilla assembled a Raspberry Pi compute module with a Basler camera and a ResNet-based model fine-tuned on 3,000 labeled images from the supplier’s own rejection bin. Tested offline against 90 days of archived inspection images, it correctly classified 91% of the escaped surface defects that had triggered the SCAR events—against the 74% catch rate the manual line had achieved on the same defect types. That gap, 17 percentage points, was the product thesis.
Qcvisionly was founded in Ann Arbor, Michigan by engineers who spent careers inside automotive tier-1 quality departments watching the same failure mode repeat. A SCAR event would arrive, a sorting campaign would begin, root cause would eventually point to a surface defect that manual inspection had escaped at line speed, and the corrective action would add another inspector to a station already running three shifts. The headcount solution does not scale and does not solve the underlying physics problem. The founders built Qcvisionly to replace that cycle with a machine learning system that learns each customer's specific defect library and scores every part at line speed without fatigue, shift variation, or lighting sensitivity. The company is currently deploying its first commercial pilot installations at stamped metal and plastic injection molding operations serving North American OEM customers with 100-ppm or tighter defect specifications.
Give automotive tier-1 quality engineers inspection accuracy that human visual inspection cannot sustain at production line speed.
Give automotive tier-1 quality engineers inspection accuracy that human visual inspection cannot sustain at production line speed.
Give automotive tier-1 quality engineers inspection accuracy that human visual inspection cannot sustain at production line speed. Every part scored at line rate. Every defect classified and stored. Every pass or fail decision traceable to a specific image, model version, and timestamp. The goal is not to replace quality engineers but to give them the data infrastructure that makes their job possible at the throughput rates modern OEM supply agreements demand. We measure our success by reductions in SCAR events, sorting labor hours, and yield holds per production window, not by software feature count or model accuracy on benchmark datasets.
Qcvisionly is at seed stage, actively deploying first pilot installations at stamped metal and plastic injection molding operations in the US Midwest. We work with automotive tier-1 suppliers that have 3–25 production lines, serve OEM customers with 100-ppm or tighter defect specifications, and have experienced SCAR events they cannot resolve by adding manual inspection headcount. That last criterion matters. We are not selling to quality teams that want to experiment with AI. We are selling to quality engineers who have already confirmed that manual inspection at their line speed is a physics problem, not a staffing problem.
Qcvisionly is a seed-stage company deploying its first pilot installations at automotive tier-1 suppliers in the US Midwest. Our current focus is validation: proving that our convolutional model fine-tuned on customer-specific defect libraries achieves lower escape rates than incumbent rule-based vision systems at the same line speed and with less maintenance overhead. We are not trying to be a full-stack manufacturing execution system. We are building one thing exceptionally well: AI-powered pass or fail decisions on stamped metal, plastic injection, and powertrain subassembly parts, integrated directly into existing PLC divert logic via OPC-UA, with complete traceability for PPAP and SCAR audit support.
The principles that govern every product decision, customer engagement, and trade-off Qcvisionly makes as a company.