Viam brings AI to the physical world through your existing devices 🎛

CV Deep Dive

Today we’re talking with Simone Kalmakis, Vice President of Engineering at Viam — a universal engineering platform for bringing data, automation, and AI into the physical world. Viam was founded by Eliot Horowitz, former co-founder and CTO of MongoDB. Simone’s own experience is deep and varied, starting out in the early days of Microsoft’s machine learning engine and continuing as a Director of Engineering at Flatiron Health and Arcadia, as well as a stint as a founder.

The common thread through her career: data‑driven intelligence that benefits society. At Viam she’s helping turn that intelligence loose on arms, cameras, sonar rigs and any other physical device that moves, senses, or operates in the real world.

Key Takeaways

• Perceive → Decide → Act in one stack – Viam installs an open-source application on edge devices that connects to hardware, ingests and processes data locally or streams it to the cloud – and allows you to interact with the hardware components – shrinking time‑to‑production from months to days to create an end-to-end solution operating in the real world.
  

• Modular by design – Hardware drivers, services, and ML inference engines are distributed as plug‑in “modules,” so swapping a depth camera or replacing a model is one configuration change, not a rewrite. Developers can publish their own modules for others to use via Viam’s open module registry.
  

• Real deployments – Early customers include marine‑sector giant Kongsberg (turning raw sonar into actionable fishing and obstacle detections) and UBS Arena (leveraging real-world data to improve crowd flow and fan experience).
  

• Hard constraints embraced – Intermittent connectivity, strict power budgets, and limited on‑board compute shape every design decision. Viam builds for environments where a satellite link can disappear without notice and computational optimization are essential.
  

• Bridging engineering silos – Full‑stack developers can program motors and sensors without hiring an embedded team; hardware specialists can drop in custom ML models or computer vision algorithms without rebuilding an app framework.
  

• Hiring across the board – Current priority roles include data engineers, robotics engineers (especially cameras and motion planning), and solutions engineers, but Viam is scaling company‑wide: high performers from any discipline should reach out. We’re a fast-paced, vision-driven company.
  

Viam raised a $30M Series C in March led by Union Square Ventures with participation from Battery Ventures, Neurone, and other existing investors, bringing their total raised to $117 million.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Simone 💬

Your journey up to Viam sounds like a tour through almost every major vertical that crunches data. What drew you specifically to a robotics platform?

I’ve always chased big missions that pair massive data with real‑world impact. At Microsoft I learned ML at internet scale, working on the core relevance engine for Bing; at Flatiron Health we used data to improve cancer outcomes; at Arcadia we used utility data to fight climate change. With Viam I get a brand‑new frontier: bringing intelligence to the physical world.

The company’s vision is huge, the technical depth is exciting (Eliot Horowitz wrote MongoDB and still contributes code here), and as an engineering leader I stay hands-on with hardware, ML, and cloud infrastructure every single week. I studied math at Yale, and I’ve put it to use more at Viam than any other company I’ve worked at. Motion planning in particular has been a very interesting new area for me. Coming from a software background, one of the most enjoyable parts of Viam is learning about a whole new domain, and our whole team is passionate about learning and teaching.

In a single sentence, what does Viam make?

A software platform that lets any developer connect to physical devices, pull their data, run intelligence on it, and push decisions back to the hardware—all through straightforward, well‑documented APIs.

Where exactly do you fit in the robotics stack?

We sit directly above your hardware layer and below your application logic. Think of us as the connective tissue: you bring your cameras, servos, sensors, or edge boards; we provide a uniform interface for collecting data, running ML models, and issuing commands. As a platform, we support a large array of hardware devices out of the gate, and with the power of our developer community, we aim to support even more. Any custom development can be added in as a pluggable module that others can reuse.

Can you give a concrete example of that “plug‑in” concept in action?

At UBS Arena all the security cameras were already installed. We dropped in our camera modules, ingested the video streams, and layered computer‑vision models on top to track crowd flow and feed real-time updates to fans. With QuickQueue in the NY Islanders app, fans can now find the shortest lines for concessions and bathrooms—no rip‑and‑replace, just new intelligence on existing hardware. If they upgrade a section from a standard IP camera to an Intel RealSense depth camera, they’d swap one module for another and keep the rest of the pipeline intact.

You mentioned marine projects with Kongsberg. How does Viam help on a fishing vessel?

Kongsberg’s sonar systems already capture dense sonar images under a boat. Before Viam, that data was hard to interpret by a non-specialist. We now ingest Kongsberg’s sonar data and apply computer‑vision models to classify shapes in the sonar image as likely fish schools for the captain to drive toward or obstacles for the captain to avoid. The result isn’t just pretty visuals—it’s actionable information that can be sent straight to the wheelhouse display or logged for sustainability reports.

What technical hurdles did you and your team face that a pure cloud engineer might never see?

• Connectivity: A vessel in the North Atlantic can lose satellite connectivity for hours. Viam must be able to run without connectivity, intelligently cache data, and handle old data gracefully upon reestablishing connection.

• Power: Many edge devices run off batteries. We empower developers to scale factors like the frequency of ML inference or video resolution dynamically to stay within a power budget.

• Compute: We run on all kinds of hardware, including devices with very little memory or processing power. Our processes are built to be compact, efficient, and fast––and to take advantage of the hardware available, such as GPUs on the Orin Nano.

Operating within these limits, we design for reliability and resource awareness from the outset.

How is AI used on the platform right now?

Almost every use case involves computer vision—object detection, crowd counting, segmentation, anomaly detection. We built a full ML workflow: ingest data from edge devices into Viam Cloud, spin up a training job on image or sensor data, and redeploy the trained model back to the edge for low‑latency inference. We recently added custom training script support, so users can bring their own architectures instead of being limited by model type. Like with hardware components, Viam is fully modular with respect to ML models as well: you can bring in external ML models and reuse models from other developers on the Viam registry.

What metrics tell you Viam has succeeded for a customer?

• Time‑to‑production – how long from “hello world” to revenue‑generating deployment.

• Downtime reduction – predictive maintenance means fewer field trips.

• Maintenance budget & support costs – correlate lower truck rolls or on‑site visits.

• Revenue generation and growth – expanded capacity means new products and bigger markets.

• Hardware coverage – how many of their existing devices they can reuse.

When a customer says, “We never could have built this solution with our existing team before Viam,” we know we hit the mark.

AI‑powered robotics is changing fast. Where does the field go in five to ten years?

Expect far less rigid pre‑programming and far more dynamic decision‑making. Robots will perceive messy, real‑world inputs and update their motion plans on the fly. Recently, to demonstrate this future vision of physical intelligence, we brought a wine-pouring robot to UBS Arena that responds dynamically to its environment in real time. Human‑in‑the‑loop will also remain important: if autonomy can handle 90% of cases and a human collaborator handles the rest, whole new sectors become economical. Voice‑guided setup, self‑calibrating arms, on‑device learning—all of that lowers the barrier so more people can deploy automation, from huge companies with no robotics experience, to a small business or homeowner.

Stretch that vision out to fifteen years. What changes for everyday life if Viam’s mission succeeds?

Smart machines become part of everyday life and fade into the background. In the home—kitchen prep, laundry folding, elder‑care monitoring—robots quietly handle chores. In industry, tasks that are dangerous, dirty, or monotonous get automated by default: inspecting high‑voltage lines or scanning sewer mains. The key is you won’t need a PhD robotics team; a full‑stack developer will install Viam, choose modules, write some code, and ship intelligent machines in days.

Who’s adopting Viam today?

We serve a broad spectrum from established enterprises to seed-stage startups that want to build on Viam from the start. On the enterprise side, we’re seeing success in sports and entertainment venues, manufacturing, marine, robotics, quick service restaurants, and more. That diversity feeds back into the platform. Features we build for an arena’s camera network can help a marine‑sonar client two months later.

How do you prioritize core platform work versus bespoke customer requests?

It’s a balancing act. Platform improvements benefit everyone, but marquee customers sometimes surface gaps that moves a farther-term project up in priority. The sweet spot is turning a bespoke ask into a reusable module or other universally beneficial feature area. 

Let’s talk about the team. Are you growing the team, and what qualities thrive inside Viam’s culture?

We’re growing across the entire company, so if you’re a high performer there’s a good chance we have a role for you. Some specific roles that come to mind right now include: 

• Data engineers to expand our ingestion, warehousing, and ML‑ops pipelines.

• Robotics engineers with deep camera or motion‑planning expertise.

• Solutions engineers who can bridge customer hardware with Viam.

What unites successful teammates is intellectual curiosity across departments. A software engineer eager to work on hardware or a mechanical engineer keen to work with code. We largely work in‑person at our New York robotics lab, surrounded by robotic arms, Raspberry Pis, and enough spare parts to build a small factory.

Why should a 10x engineer pick Viam over another hot startup?

Two big reasons. First, the ambition: turning every physical device into a smart, connected system is a generational challenge. You’ll never run out of green‑field problems. Second, the chance to learn directly from Eliot Horowitz, former co-founder and CTO of database giant MongoDB, where he built the technology integral to their success and IPO. He’s in the Viam code every day and constantly building on the platform, and that hands-on culture permeates the whole org.

One leadership insight you wish every engineer internalized early?

Strategy isn’t just for VPs. Knowing why a feature moves the business helps at every level—whether you’re designing an API or where to draw the line for an MVP. One of my staff engineers checks in with me monthly with the question “What’s the highest‑impact project for the company right now?” That question keeps our work aligned and our impact optimized. I encourage engineers at every level to adopt that mindset. Understanding why you’re building what you’re building helps you build better.

Conclusion

Stay up to date on the latest with Viam, learn more here.

Read our past few Deep Dives below:

• Arcade - A New Kind of Marketplace 🔍

• Strong Compute - Your High-Performance Training Infra Platform 🔋

• V7 Go - AI Agents for Complex Document Workflows 🎛

• Zencoder’s Repo Grokking - Your AI coding unlock 🔑

• Rime - Authentic AI Voices for Real-time Conversations at Scale 🗣️

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

Join Slack | All Events | Jobs