Overview AI launches OV Spark line of AI inspection cameras

Overview AI today launched the OV Spark and OV Spark Pro. The new cameras make up a new line of AI inspection cameras with an agentic AI layer that answers questions in plain English.

Spark ships with 10 inspection tools that can be mixed and matched in a single recipe. Seven are deep learning AI models: AI Trigger, Alignment, AI Classification, AI Segmentation, AI Count, AI OCR, and Anomaly Detection.

The other three are ready to inspect out of the box, with the option to add images when an inspection needs training. This includes for barcodes, measurements, and color match.

“In a robot cell, the part doesn’t always land the same way twice,” Christine Chacko, a technical product marketing manager at Overview AI, told The Robot Report. “OV Spark anchors on features of the part itself, so every inspection region follows it as it shifts or rotates.”

“If no part is found, the capture fails rather than passing an empty frame,” she explained. “It holds 10 inspection tools, and the PLC can select the recipe by ID as the arm moves between stations. Adding a station thus requires only a new recipe number, not another camera.”

The recipe, model training, inference, judgment, PLC communication, image library, and the operator interface all continue to run on the camera itself.

“We have been putting AI on factory floors since 2019, and the hardest part isn’t the inspection itself. It’s that every new line means another project, another specialist, another few weeks,” said Chris Van Dyke, co-founder and CEO of Overview AI. “Our mission is to catch every defect in every factory, so the camera has to carry that setup work itself. Spark is our best product at doing exactly that.”

Overview AI said that OV Spark and OV Spark Pro are available now directly from the company and through its distribution partners.

Why did Overview AI create Sparky?

Despite improvements to the accuracy of its vision models, Overview AI said customers still ran into a gap in consistency. Which tool to use, and what to take from the production data, still came down to individual judgment. This led Overview AI to create Sparky, the AI agent that lives on the Spark camera.

Sparky gives teams a shared starting point for these judgment calls, said the company. This enables everyone on a team to understand reasoning behind an inspection in the same way.

“The onboard agent tells the operator which tool a check needs and why. Then Sparky answers questions about any capture in plain English,” Chacko said. “A robot cell can produce thousands of captures that nobody ever reviews. An operator can prompt Sparky to understand what failed in a given timeframe and receive data that helps to retain the model for the next set of inspections.”

When it comes to agentic tool selection, Sparky recommends the tools for an inspection and explains the reasoning behind each one. If asked about a connector that must be confirmed to have 24 pins before it goes to a PLC, Sparky explains that the Alignment tool is needed to follow the part as it drifts, and the AI Count tool is needed to confirm that every pin is accounted for.

For agentic diagnostics, Sparky makes the capture library searchable in plain language. This means a question about the previous shift returns a timeline of the captures, each with its verdict and the tool results associated with it.

As an agentic human-machine interface (HMI) builder, Sparky lays out the operator screen from a description of the data a station needs, using widgets built to show product metrics at a glance.

“For years, manufacturers have told us the same thing: they have more inspection points than they have people qualified to set them up. Spark is our answer to that,” said Russell Nibbelink, co-founder and chief operating officer of Overview AI. “A quality engineer who knows the part but has never trained a model can now get an inspection running on the line, and that changes who inside a plant can actually do this work.”

OV Spark features new hardware

The 1.2MP OV Spark and the 5MP OV Spark Pro share the same sealed, all-in-one body, rated IP66 and small enough at 70 x 116 x 53 mm (2.7 x 4.5 x 2 in.) to drop into existing cells. The Spark can be powered over Ethernet through a PoE coupler.

“We took seven years of what we learned deploying AI inspection in real plants and compressed it into one camera: Spark,” said Xiaoyang Kao, co-founder and chief technology officer of Overview AI. “We designed it on Qualcomm Dragonwing and put the whole agentic stack on the device. The camera recommends tools, reads production data, and lays out the operator screen itself. We’ve always run intelligence at the edge. Spark is the first product where that intelligence can support an expert in doing their job.”

Mansfield Engineered Components uses OV Spark

The OV Spark is already running at Mansfield Engineered Components (MEC), a metal fabricator and contract manufacturer in Mansfield, Ohio. There, the camera is on an end-of-line station that had no formal inspection before it, only an operator’s visual check.

A missing washer is the most frequent defect on that line, and the inspection was built by a quality engineer who had never worked with a vision camera before joining the company a year ago. MEC has since moved the camera between two lines and is scoping a second Spark for a rivet curl and stenciled date code check.

“I’m a quality engineer with an industrial engineering degree, about a year into the job, and I had minimal experience with cameras before this,” said Logan Hoon, the quality engineer at MEC who built the inspection, on what it took to get the station running.

“The part I expected to be difficult was the aligner, and it turned out to be pretty easy, because [the camera] tells you exactly what to do,” Hoon said. “I think we knocked it out of the park. [Spark is] reliable, easy to use, and easy to train your maintenance people on.”

---

---