Your cameras are already there.
Now make them think.

Walk into the security office of a mid-sized warehouse in Indiana today, and the setup looks roughly the same as it did in 2004. Banks of monitors. Grainy footage. A bored guard scanning feeds manually. If something happened Tuesday, good luck finding it by Thursday. The camera captured it - somewhere in 800 hours of footage nobody has time to scrub.

This is the moment Coram AI was built for.

The Sunnyvale-based company has built an AI video security platform that plugs into whatever IP cameras a business already owns and transforms them into a searchable, alerting, incident-managing system. No ripping out hardware. No six-figure deployment. Just a software layer that makes the cameras people already paid for do the job they were always supposed to do.

The physical security market is enormous - estimated above $30 billion globally - and it is, to put it diplomatically, not known for technological ambition. Most enterprise security setups are a patchwork of different camera brands, aging NVRs, and software that was designed when "the cloud" was still a PowerPoint metaphor.

The result: footage gets captured, stored, and almost never usefully retrieved. Incidents are detected after the fact, if at all. Security teams spend more time clicking through timelines than they do actually preventing anything. And the price of switching to something better? Usually a full hardware replacement that costs as much as the problem it's solving.

The market had resigned itself to this. Legacy vendors had optimized for selling more hardware. Nobody had asked whether the software could just be smarter.

Ashesh Jain and Peter Ondruska spent years building systems that let cars see and understand the world around them. Jain led autonomy and AI at Lyft's self-driving division, holds a PhD from Cornell, and was a visiting scholar at the Stanford AI Lab. Ondruska has a PhD in robotics from Oxford, and before joining Lyft sold an augmented reality startup to the company in 2018.

When they left in 2021, they could have done almost anything with those credentials. They chose to fix security cameras. It's the kind of career pivot that sounds eccentric until you see the market size - and until you realize that the computer vision problems powering self-driving cars and the ones needed to make surveillance useful are, technically, not that different.

The flagship capability that turns heads is natural language video search. Type "person in a gray hoodie near the east exit between 2pm and 4pm Tuesday" and Coram finds it. Not by matching a timestamp you already know - by actually understanding the query and scanning the footage. It's the kind of feature that sounds like a demo trick until a security manager at a school district realizes they've just saved four hours of investigation time.

But search is just the front door. Behind it is a platform that handles real-time threat detection (firearms, slip-and-falls, PPE violations), facial recognition with known-offender alerts, license plate tracking, access control integration, emergency management with wireless panic buttons, and visitor management. All of it in a single dashboard. All of it running on cameras the customer already owns.

Coram's customer base spans a curious range: Fortune 500 corporations alongside a church in Washington state, K-12 school districts running active-shooter detection alongside grocery chains watching for slip-and-falls. The common thread is that all of them already had cameras, and all of them found the existing software inadequate.

Middletown Unified School District in California adopted Coram for campus safety monitoring. PCC Community Markets deployed it across grocery locations in the Pacific Northwest. A statewide Indiana government contract in February 2026 signaled that public-sector buyers are paying attention.

G2 users gave the platform a 4.9 out of 5, which in enterprise software circles is the equivalent of getting a standing ovation. The ease-of-use score of 9.5 out of 10 suggests that all that PhD-level engineering actually made things simpler for the people who have to use it daily.

Battery Ventures led the Series A not because they were looking for a flashy AI play, but because they understood that physical security infrastructure is a massive, underserved market with decades of technical debt. Marcus Ryu, the Battery partner who joined Coram's board, previously co-founded and led Guidewire Software - a company that did exactly the same thing for insurance software that Coram is trying to do for security. The pattern is deliberate.

The mission Coram has staked out is not modest: transform physical spaces into intelligently managed environments. Every building is already instrumented with cameras. What's been missing is the software layer that makes all that recorded data actionable - in real time, not after the incident report is filed.

The company's focus on working with existing infrastructure rather than replacing it is both a commercial strategy and a genuine technical position. Ripping out cameras is disruptive, expensive, and environmentally wasteful. Making them smarter is better on every dimension. It also lowers the barrier for adoption in exactly the environments that need security most: underfunded school districts, community hospitals, small-city governments.

There's also a privacy angle the founders take seriously. Coram's architecture is designed to give organizations useful intelligence without turning their facilities into surveillance states. End-to-end encryption, HIPAA compliance, and granular permission controls aren't afterthoughts - they're built into the platform from day one.