Scaling edge AI from Pune: Inside IndoAI’s big bet on smart cameras
With in-house hardware, a vision model marketplace, and a growing client base, IndoAI aims to capture 35% of the global AI camera market in the next four years.
During the COVID-19 lockdown, Rashmi Kulkarni and her team built a mobile app to mark student attendance using facial recognition. When users began tricking the system with videos and images, she and Co-founder Eric Fonseca realised the need for a more secure, scalable solution.
In 2021, that challenge gave rise to Pune-based startup IndoAI, which develops AI-powered edge cameras that run multiple vision models in real time without relying on external servers.
CEO Kulkarni has worked at TCS and later at SpiderG. Fonseca, Kulkarni’s classmate from DY Patil College, started his career in quality analysis before joining a fintech project called SpiderG.
“We knew each other from college, and after working in different domains, we came back together,” says Fonseca, who now manages IndoAI’s marketing operations.
Together, they launched their first product, DutyPar, which works by using facial recognition on mobile devices to mark attendance in real time, ensuring touchless and forgery-free verification. By the end of the year, it had shifted the focus to edge AI cameras.
Building hardware in-house
Unlike most vision-based startups that rely on existing CCTV systems and external servers, IndoAI develops its own hardware. The company uses Nvidia ARM processors like Jetson Nano and Orin to power the devices, which process video on the edge rather than sending raw footage to the cloud.
The system is useful in places like government offices, hospitals, or remote areas with poor internet or strict data rules. “We process everything on the edge and send live alerts,” Kulkarni says.
The cameras also have power backup, SIM support for low-network areas, and an integrated application called IndoAI, available on web and mobile, which sends alerts in stages—first a push notification, then an SMS, and finally a call if no one responds.
How it works
IndoAI’s flagship AI edge camera is built to run multiple vision models simultaneously. A basic variant supports one or two models, while higher-end units can run four or five. “If today anyone wants fire and smoke detection, and tomorrow intrusion detection, they can install it with one click,” Kulkarni says.
The AI models are also industry-specific. It also offers facial recognition and body gestures, so a jewellery shop may use the product for theft or chain-snatching detection, schools can run attendance and unauthorised entry models, and housing societies can track vehicles and visitors.
The company is also building what it calls an “Appization” (AI marketplace). “It’s like a Play Store for AI models,” Fonseca says. Customers can install models over the air, while independent developers can contribute their own, test them live, and share revenues.
IndoAI manufactures entirely in Pune. Its team of 35 includes 20 tech employees. The cameras’ casing, software, and large language model (LLM) are developed in-house in partnership with IIT Delhi and IIM Lucknow incubators. “We don’t use any Chinese components; it’s completely made in India,” Kulkarni tells YourStory.
For AI training, the startup collaborates with IITs and IT colleges to build datasets and partners with firms like Seven Sense. “The LLM is trained on custom datasets specific to use cases like intrusion or theft,” Kulkarni says.
Customers and the market
The company follows the B2B2G model, serving both enterprises and government bodies, while also reaching smaller clients like jewellery shops and cafés. “That’s why we built a variant that can run only two models, making it affordable,” Fonseca says.
IndoAI’s first client was the Maharashtra State Skill Development Department, which used its DutyPar app during the pandemic. Since then, the startup has deployed over 425 cameras in schools, hospitals, housing societies, cafés, factories, and gram panchayats in Andhra Pradesh and Madhya Pradesh.
It also works on smart city projects in Andhra Pradesh, Madhya Pradesh, and Maharashtra, with features like vehicle tracking, mob detection, and women’s safety alerts.
Revenue model and pricing
IndoAI earns through three streams: DutyPar, its AI marketplace, and camera sales. The cameras cost between Rs 1.5 lakh and Rs 5 lakh, depending on capacity and features.
“For special needs like on-premise processing or system integration, we charge more,” Kulkarni says.
Unlike SaaS products, IndoAI runs on a “pay-as-you-go” model. For clients who do not want to share video feeds externally, all data is processed on their premises.
The startup reported Rs 64 lakh in revenue in FY24 and expects to cross Rs 20 crore in FY26. It has raised around Rs 7 crore in seed funding from friends and family, and is now looking to raise at least Rs 50 crore to meet a strong order pipeline worth Rs 15 crore.
Standing out in a crowded market
Global players like Axis and Vision AI also make smart cameras, but Kulkarni believes IndoAI stands out by processing on the edge and running several models on one device.
“Other cameras usually handle only one model at a time. For multiple uses, you need multiple cameras,” she explains.
The team also hopes its marketplace will drive growth. “We want to be the Android of AI cameras,” Fonseca adds.
Since IndoAI manufactures all its hardware in-house, the biggest hurdles lie in sourcing components and scaling production. “Hardware is hard to source, and scaling production is not easy,” Kulkarni says.
Market size and the road ahead
According to the Grand View research report, the global AI camera market, which includes smart cameras like those developed by IndoAI, was valued at around $13.9 billion in 2024 and is expected to grow to about $47 billion by 2030, with a compound annual growth rate of approximately 21.6%.
“We’re hardly capturing around 1%, but in the next 3-4 years we’re aiming to capture 35%,” Kulkarni says.
Looking ahead, IndoAI plans to expand its product line, onboard more developers onto its AI marketplace, and grow its government and enterprise clientele.
It is also working on a women’s “safety gesture” model, which will allow a simple hand movement to trigger alerts to the nearest police station or administrator. “We’re at 60–65% accuracy, but we launch only after 95%. The rollout is planned for April 2026,” Kulkarni says.