There are now more cameras in the world than people, and most of them just sit there recording. Atul Rai, co-founder and CEO of Staqu Technologies, set out to make them think.
In this episode he explains video analytics AI: how Staqu pivoted from fashion image search, why it started with the police, and how its Jarvis platform now serves retail, manufacturing and more.
What is video analytics AI? Atul Rai’s core thesis
Atul’s view is that CCTV cameras are mostly passive devices that only help after something has happened. Staqu adds an AI layer so the same cameras can detect events, count people, raise alerts and answer questions.
Computer vision is the underlying field, and Staqu’s founders came to it from research, not business, which shaped how they built and sold.
- Make existing cameras smarter. Software sits on top of cameras companies already own, rather than replacing hardware.
- Earn credibility in a tough market first. Police deployments built trust that opened doors in retail and manufacturing.
- Be honest on accuracy. Staqu sets expectations at about 95 percent and pilots quickly.
- Software is India’s strength. Atul argues Indian teams can win on tech even where hardware depends on imports.
How did Staqu go from image search to video analytics?
Atul studied computer science and began working on image processing during his B.Tech, later publishing research and completing a masters in AI at the University of Manchester. After research work in Europe, he returned to India and founded Staqu in 2015 with two co-founders who brought engineering and back-end strength.
The company first built image search for e-commerce, such as finding a product from a photo. As e-commerce consolidated and the team realised its algorithms could also recognise faces and appearances, they pivoted to video analytics.
- Started with fashion and visual search
- Reused the same recognition tech for appearance and face analysis
- Targeted CCTV, where huge volumes of video go unused
Why did Staqu start with police forces?
Cameras are personal and security-driven, so Atul felt retailers would not trust an unknown startup with them. Working with police forces first gave credibility, and Punjab was chosen as a tough test for facial recognition because of beards and turbans.
After a successful pilot, other states adopted the technology, largely through word of mouth between police departments. With that credibility, Staqu went back to retail and manufacturing.
What problems does video analytics solve in retail and manufacturing?
Atul splits use cases into loss prevention and revenue generation. Loss prevention covers security and safety for government, infrastructure and manufacturing, while revenue generation is mostly retail.
Security and safety
Examples he gave include intrusion detection, crowd build-up, left baggage, violence, fire detection, safety gear checks and kitchen hygiene audits for food-tech companies.
- Intrusion and crowd alerts
- Helmet and safety jacket checks
- Fire and hygiene monitoring
Retail insights
Retailers know sales but not how many visitors came or what they looked at. Footfall counting, queue management, heat maps and planogram analysis fill that gap without intrusive methods.
- Footfall and queue length
- Heat maps of store movement
- Insights on which products get attention
What is Jarvis and how does it work?
Jarvis is Staqu’s platform that adds analytics to existing cameras. Atul describes it as plug and play: the customer copies the camera’s IP address and starts using it, or installs a local app to access the camera feed.
Staqu also built image enhancement to improve blurry frames before prediction. Where possible, the AI is deployed on the customer’s own server, which Atul presents as a privacy advantage over a purely cloud model.
- Connect the camera feed through its IP address or a local app.
- Enhance low-quality frames if needed.
- Run detections for activities, classification and alerts.
- Review alerts and insights on the platform.
How does Staqu handle accuracy, data and expectations?
Atul says accuracy depends on camera quality, lens and bandwidth, so Staqu tells customers upfront what is realistic, around 95 percent, and runs short pilots of a few days. He says overpromising is the wrong approach.
On data, the company uses an internal curation team to tag data and train models, and notes the importance of data protection rules.
What business lessons did a research-led team learn?
Atul says the founders were techies first, so they had to learn the business. His main lesson is to stay close to the customer, even spending time on site to see where the real problem is.
He also describes a mix of private and government revenue, roughly 60 percent private and 40 percent government at the time, and a push into the Middle East and London through direct sales and channel partners.
Further reading and sources
- Computer vision (Wikipedia)
- Closed-circuit television (Wikipedia)
- Video content analysis (Wikipedia)
Frequently asked questions
What is video analytics AI?
It is software that analyses video from cameras to detect events, count people or objects and trigger alerts. Atul Rai describes Staqu’s approach as adding a smart layer to existing CCTV cameras rather than replacing them.
Can video analytics work with the CCTV cameras I already have?
According to Atul, yes. Staqu’s Jarvis is plug and play: you connect the camera’s IP address or use a local app. Results still depend on camera quality and bandwidth, so he recommends a short pilot.
How accurate is AI video analytics?
Atul says accuracy varies with camera hardware and network, so Staqu tells customers to expect around 95 percent rather than promising perfection. A short pilot on your own cameras shows what is realistic.
Conclusion
Atul’s story shows how a research-led team can find a commercial market: pick a hard problem, earn trust where credibility matters most, then expand.
If you are building a deep tech product, run a quick pilot, be honest about accuracy and stay close to your customer.
Want more conversations like this? Join The Builders Club, our community of 35,000+ founders and CXOs, and explore more episodes of the Startup Founders Podcast.
