A new Seattle startup wants to make it easier for businesses to monitor their warehouses using artificial intelligence and computer vision technologies.
Groundlight emerged from stealth Wednesday, unveiling its platform that can enable image understanding from natural language queries and a few lines of code. The company also announced $10 million in seed capital.
Launched in October 2020, the startup was co-founded by Leo Dirac and Avi Geiger. Dirac worked as a senior principal engineer at Amazon Web Services for more than six years, where he led the AutoML program.
Geiger is a longtime product design and engineering consultant. He’s also the former CTO and co-founder of high-tech appliance maker Picobrew.
The duo teamed up to figure out how to reduce the time and money required to create machine learning applications. The process typically involves hiring a team and a lengthy process of data gathering, labeling, training and iteration, Geiger told GeekWire.
Large language models such as OpenAI’s GPT-4 are powerful but have limitations, being often too slow and expensive for industrial use. They can also be confidently wrong, so systems are needed to ensure accuracy.
Groundlight offers a unique solution for building computer vision systems using natural language instructions tailored to specific needs that allows for instant customization.
The technology taps into a company’s network of cameras to gather real-time data. From there, users can input prompts — such as, “is there a truck in the loading dock?” — and then get results instantly.
Groundlight’s approach differentiates from conventional computer vision and machine learning operations solutions because of its multi-layer system, according to the company.
If the model has low confidence in a result, it redirects the query to more advanced machine learning systems, and escalates to a human specialist if needed.
Groundlight offers 24/7 human monitoring for queries that are too complex or ambiguous for its machine learning models. This approach helps to limit the amount of errors, Dirac said.
“A key point of discussion in the tech community right now is about trust of these systems,” he said. “Our system is architected from the beginning to always have humans in control.”
The outcomes to the questions, produced by either human or machine, are fed back into the model. This lets the system respond to future queries more efficiently.
The model is designed to handle binary queries — yes or no questions. The queries are ultimately something that a human could answer, Geiger said.
Groundlight’s platform can be used in a variety of applications including video stream analysis, industrial automation, process monitoring, retail analytics, and robotics.
Groundlight’s launch comes as a growing number of companies look to automate different functions in manufacturing, as well as a push to onshore production processes and an ongoing labor shortage.
Austere Manufacturing, which produces cam buckles, uses Groundlight’s API to inspect its products and monitor processes. The company said all it needed was a $10 camera and a few lines of code to set up the system.
There are a number of startups selling AI-powered tools to manufacturing companies, including Seattle-based Loopr, which offers edge-based software that can detect defects in the products.
Groundlight’s funding round was led by Madrona Venture Group. Other investors include Greycroft Partners, Founders’ Co-op, Flying Fish, Ascend, and Essence VC.
“Every company has unique data, especially in industrial applications,” Madrona Managing Director Tim Porter said in a statement. “Groundlight’s approach enables companies to utilize off-the-shelf cameras and inexpensive equipment to quickly build and reliably operate customized models.”
In a blog post, Porter said Groundlight “has the potential to become another iconic software and AI company based in the Pacific Northwest.”
“Avi and Leo’s vision to make high-quality [computer vision] as simple as integrating an API service like Twilio resonated deeply with us,” he wrote.
