Moii.AI: Deploying vision AI agents on CCTV to improve safety and productivity with BigQuery and Gemini
2+ days of manual video scrubbing reduced to real-time automated detection
$10k saved per month on AI processing with the Google for Startups Cloud Program
Doubled engineering team efficiency by using out-of-the-box Google Cloud infrastructure solutions
Moii.AI uses Google Cloud infrastructure to help its customers deploy vision agents and drive real-time actions that improve their bottomline.
Moii.AI uses Google Cloud infrastructure to help its customers deploy vision agents and drive real-time actions that improve their bottomline.
Using AI to make sense of CCTV camera data
Google Cloud provides the ideal framework for AI enablement. Its end-to-end tools allow us to prioritize building new solutions for our customers rather than worrying about whether or not our infrastructure works.
Deepak Upadyaya
Head of Product and Co-founder, Moii.AI
The internet is information dense and overflowing with data, but there's also a substantial amount of data in physical spaces. With more than a billion CCTV cameras in the world, there's a vast amount of video footage that contains vital information about workplace safety and productivity, but sifting through that footage can take hours or days to uncover useful information. Moii.AI saw an opportunity to automate this review, analyze, and action process with AI and turn traditional camera systems into autonomous agents to create safer and more productive workplaces.
Tapping into this video data requires computational and data infrastructures that can process petabytes worth of information quickly to even be functional.
To reach a state that the data is useful, the company must also parse, process, and label data as well as be able to share updates with end users who can take action on that information.
Moii.AI also needed this solution to be configurable for different customer needs without having to build them from the ground up. For example, a security team at a large organization may need to be able to scan footage for specific vehicles entering or exiting their facility, but a maintenance team may rely on its cameras to locate potential slip hazards. "We're focusing on building context models, so we don't have to retrain our solution from scratch for every possible use case," says Lakshman Balasubramanian, Head of AI and Co-founder of Moii.AI.
This focus led Moii.AI to Google Cloud and Gemini for developing robust, flexible data sets quickly. "Google Cloud provides the ideal framework for AI enablement," says Deepak Upadyaya, Head of Product and Co-founder of Moii.AI. "Its end-to-end tools allow us to prioritize building new solutions for our customers rather than worrying about whether or not our infrastructure works."
Processing petabytes of video data in real-time
The company started its journey with Google Cloud using Firebase to build the initial version of its application, and as it began to expand, it saw more and more solutions to create its cloud infrastructure. "We started with Firebase, but every time we looked for a new solution, Google Cloud had the exact infrastructure product we needed," says Upadyaya. "The freemium model gave us the opportunity to test more and more, and when Gemini released, we knew Google had everything we needed for both our infrastructure and our vision AI pipeline."
Moii.AI found that Gemini's two million context window provided the bandwidth necessary to process the large volume of data required to analyze hundreds of hours of video footage.
With BigQuery, video processing and review is near real time. What used to take five people manually scrubbing through footage for two or three days is now a five-minute search.
Lakshman Balasubramanian
Head of AI and Co-founder, Moii.AI
Beyond the technical components, the team also learned that through the Google for Startups Cloud Program, it could access credits that would save them $10k per month on the solutions it was already testing.
Along with building and maintaining its own infrastructure, Moii.AI uses Google Cloud solutions to implement on-premise applications and storage for its customers with Google Cloud Storage. These systems capture footage, break it down into two to three minute segments, and create metadata to describe and categorize objects in the video.
Segments captured on-site are then processed by Moii.AI's AI and machine learning models. "Our object detection models work with Gemini 1.5 Pro, Vertex AI, and our other Visual Language Models (VLMs) as we ingest customer footage to augment the work the on-prem models already do," says Balasubramanian. These redundancies ensure Moii.AI is training its models to accurately detect certain objects before feeding data into BigQuery. Once there, footage is searchable based on metadata keywords, so even non-technical users can search for and access relevant footage.
With Pub/Sub, Moii.AI can also directly push notifications to end users who can be notified based on specific events, such as cameras detecting an employee entering a cordoned-off area of a warehouse floor or someone using a phone while operating heavy machinery. "With BigQuery, video processing and review is near real time," says Balasubramanian. "What used to take five people manually scrubbing through footage for two or three days is now a five-minute search."
This accelerated review process can help Moii.AI's customers save hundreds of hours of labor and more effectively avoid potential risks, saving organizations millions of dollars. The team has also been able to standardize this deployment process and consistently use the same solutions across customers, making it easier and more efficient to deploy as the company reaches new customers.
Designing a scalable, reliable AI infrastructure for a rapidly growing company
Because of BigQuery’s intuitive UI, even engineers without a lot of cloud experience can go into the tool in their first week and start to work on data projects. With another solution, we would need a team twice this size to maintain the infrastructure we’ve built.
Lakshman Balasubramanian
Head of AI and Co-founder, Moii.AI
Beyond its cutting-edge customer-facing solutions, Moii.AI is also building a complex, scalable cloud infrastructure behind the scenes. "I'm grateful that we don't have to worry about our infrastructure being a limitation," says Upadyaya. "We've never had an outage, and trusting the reliability of Google Cloud lets us focus our resources on our AI models and our growth."
Moii.AI has also been able to manage this growth with a lean engineering team because it's able to assign cloud and data infrastructure work to even junior engineers. "Because of BigQuery's intuitive UI, even engineers without a lot of cloud experience can go into the tool in their first week and start to work on data projects," says Balasubramanian. "With another solution, we would need a team twice this size to maintain the infrastructure we've built."
With Google Cloud Monitoring, the team has complete visibility across its infrastructure to determine what resources are being used and where. This means it can also make decisions to spin up or spin down resources as necessary, such as during a new customer deployment.
As Moii.AI looks toward offering new solutions to more organizations, it plans to train its own CCTV-specific VLMs using Gemini Pro and Google Cloud GPUs and TPUs for processing. With these models, the company plans to automate even more of the footage annotation process, so employees only have to review relevant videos. It also wants to take it a step further by transforming the search function into a conversational chatbot and building a truly customizable, real-time video analysis platform.
Moii.AI is an innovative software startup founded by two Broad Spartan alumni, Madhu Posani and Deepak Upadyaya.
Moii.AI's main idea is to utilize Vision AI to drive real actions that improve people's daily lives. The technology framework is flexible enough to tackle problems related to operational efficiencies, residential comforts, and security. The self-learning AI is modeled after automotive technology for self-driving cars, detecting objects, and understanding movement behaviors and patterns.
Industry: Technology
Location: United States
Products: BigQuery, Gemini, Google Cloud, Vertex AI, App Engine, Cloud Monitoring, Cloud Storage Cloud Run, Cloud SQL, Compute Engine, Firebase, Pub/Sub
