When Ganesh Ramalingam was young, his family got a phone call that changed everything. His uncle had died in an industrial accident: sudden, unrecorded, and by most measures, preventable. That call stayed with Ganesh for years.
“One accident changed everything for our family. That never left me,” he said.
Zapdos Labs is what came out of it.
A Problem Hidden in Plain Sight
Most people who have never worked in a factory would be surprised to learn how little real-time safety monitoring actually exists on the average plant floor. The United States alone has more than 250,000 manufacturing establishments, and similar numbers across Southeast Asia, the vast majority of them tier 2 and tier 3 plants. These mid-market facilities employ a significant share of the industrial workforce, yet have long been priced out of the safety technology available to major corporations. Legacy computer vision systems require months of labeled training data, large technical teams, and capital that smaller operations simply do not have.
The result is a gap that gets filled with paper checklists and periodic human audits. According to the company’s customer interviews, facilities can spend an average of $400,000 per year on compliance while carrying approximately $7 million in unhedged liability from near-miss incidents that were never recorded. Multiplied across the global mid-market footprint, that is tens of billions of dollars in annual compliance spend generating almost no continuous digital record. Workers go home, and very little of what happened that day exists anywhere in digital form.
Before starting Zapdos Labs, Ramalingam served in an elite signals intelligence unit, where reading patterns out of raw feeds — radio, sensor, video — was the daily discipline. That instinct sits at the core of what he went on to build with his co-founder, Tri Nguyen, who joined as CTO after leading engineering at Ubisoft and holds a Master’s in machine learning from the University of Basel. Video, they argued, could now be read the way text is: processed, understood, and acted on in real time. The cameras were already there. What was missing was the software layer on top of them.
From Safety Manual to Live Monitor
Zapdos Labs builds AI Video Agents that connect to a factory’s existing CCTV cameras without requiring new hardware or a lengthy installation. The software then reads the facility’s safety manuals and configures itself accordingly. The system identifies the specific rules of that plant: where workers should not stand, what protective equipment is required, and which behaviors signal a near-miss rather than routine activity.
Once live, it monitors camera feeds continuously, using the same kind of reasoning a large language model applies to text, but applied to video. When a risk is detected, alerts are sent instantly via Telegram, WhatsApp, or the facility’s existing video management system. The whole deployment takes approximately two days.
What separates the company from competitors is the absence of a data requirement at setup. Existing rivals in the space still require manual configuration and labeled training data for each use case. Zapdos reads a standard operating procedure document and self-configures, with no labeling and no waiting.
“Video AI has finally caught up with what manufacturers actually need,” said Nguyen, the company’s CTO and co-founder. “We chose to build on vision-language models because they let us learn each customer’s environment from documents the customer already has. No labeling, no waiting. That changes who can afford continuous safety monitoring, and over time, what else we can run on top of the same infrastructure.”
Sixty Days, Enterprise Pilots, and a Defense Contract
Traction in enterprise software usually takes time. Sales cycles stretch, pilots extend, and procurement moves slowly. Zapdos Labs did not follow that pattern. Within 60 days of going live, the company had signed Fortune 500 manufacturing pilots and a contract with the Air Force.
For a company still in pre-seed territory, that pace was unusual enough to draw attention. Zapdos Labs has since closed a $500,000 USD pre-seed funding round and is actively deploying across anchor sites in Texas and Singapore, with further deployments rolling out across North America and Southeast Asia.
The recognition has come from multiple directions. Slingshot 2025, Enterprise Singapore’s national deep tech competition, named Zapdos Labs among its Top 60 DeepTech Startups amongst 8000 companies. On the open-source side, Unblink, the company’s video infrastructure project, has accumulated more than 1,400 GitHub stars, a signal of developer-community credibility that enterprise contracts alone cannot manufacture. The team was also accepted into the NVIDIA Inception Program, placed first at the Nebius Hackathon for Real-Time Vision Language Models, and received an ElevenLabs grant.
The longer ambition is bigger than safety. When banks got software, it became fintech. When retail got software, it became e-commerce. Factories have been waiting for their turn. Over a billion cameras are already watching the industrial world. Tens of millions are inside factories. Almost none of them are reading it. Zapdos is the reader. Safety is just the first thing it looks for.
“The mission is simple,” Ganesh said. “Make sure no family ever has to take that phone call again.”
