NewWho’s building in Taipei. See Taipei’s startups, and add yours.Explore
Georgi
INDEPENDENTLY AUDITEDISO 27001SOC 2 Type II

Edge AI for
the physical world.

Georgi builds multimodal stacks for edge AI, so devices can make sense of what’s happening around them, as it happens.

Most of the world
happens offline.

Most devices just collect data. Then it all goes to the cloud to be analysed: slow, costly, and the data leaves the device. Or nobody looks at it at all. What if the device could make sense of it on its own?

Data, understood on the spot.

Edge AI runs on or near the device itself, instead of in a faraway cloud, so it can respond the moment things happen.

CLOUD AIEDGE AI · GEORGI
SPEEDCLOUDSends data to a server, then waits for the answer.GEORGIDecides on the device, the moment it happens.
CONNECTIONCLOUDStops working when the network drops.GEORGIKeeps working offline.
PRIVACYCLOUDRaw video and audio leave the device.GEORGIRaw data stays on the device. Only results are shared.
COSTCLOUDConstant streaming and cloud computing.GEORGILess bandwidth, lower cloud costs.

One stack, across chips.

Putting AI into a device is still slow and fragmented: a different tool for each chip and each sense, and AI expertise to stitch them together. Here’s how Georgi fills the gap.

01 · BUILDInference model

We build a compact model for one job, then quantize and optimize it for inference within a device’s memory and power.

02 · ADAPTChips

We adapt the stack to each chip, so it plugs in alongside your chips and your device.

03 · SHIPDevices

Device makers put it in their cameras, sensors and appliances, without an AI team of their own.

Building a smart device?

Whether it’s a camera, a sensor or a connected appliance, add AI without building an AI team, and without being locked to one chip.

Portable

One stack, many chips. Switch chips without starting over.

Small

Even a smart light bulb runs on a tiny, low-power chip. Our goal: devices like that can respond to what’s around them.

Open

What the device sees can plug into the AI tools people already use, through open standards.

A partner for chip makers.

Georgi adapts its stack to your chips,
so builders on your hardware get AI that’s ready to use.

  • Tuned to your hardwareThe stack is adapted to your NPU and toolchain.
  • A dedicated model teamAn independent team focused on models, so you don’t need one.
  • Raw data stays on the deviceOnly what the model finds is shared.
  • Built in TaiwanRight next to the chip and hardware makers.

See edge AI at work.

Our focus is the stack. The Georgi Device shows what a device running it can look like, and the stack is built to work with other makers’ chips and devices too.

Q1 2027
Georgi Device

Built on NVIDIA GB10 to run Georgi’s edge AI stack. It connects cameras, sensors and business systems, and gives every site a memory: what happened, what was decided, and what worked.

Bring AI to your devices.

Tell us a little about your company, and we’ll get back to you.

Common
questions.

What is Georgi?

Georgi is building multimodal stacks for edge AI, so builders can put AI into everyday devices in the physical world. One stack brings sight, sound and context together, and plugs in alongside your chips and your device.

What is edge AI?

Edge AI is AI that runs on or near the device itself, like a camera, sensor or speaker, instead of in a faraway cloud. The device can understand what’s happening around it and respond on the spot.

What is a multimodal stack?

It’s the set of tools that lets a device make sense of more than one sense at once: what it sees, hears and senses. Instead of a separate tool for each, Georgi brings them into one stack that works across chips.

Who is Georgi for?

Builders, founders and startups putting AI into devices, especially anyone building hardware or sourcing it in Taiwan. Chip makers work with us as partners, so the stack runs well on their hardware.

How is this different from other edge AI tools?

Most edge AI tools handle one chip family or one task at a time. Georgi brings sight, sound and context together in one stack, so builders don’t have to stitch it up themselves.

What happens to the data?

With edge AI, raw video and audio don’t need to leave the device. What the stack finds can be passed to other AI systems through open standards.