AI
The Problem Argus Solves
Argus opened its Build demo by identifying a Red Bull can and a coffee cup on voice command, then reading a room number, two names off a door, and a label on a 200mg ibuprofen bottle. See the Argus live object-identification demo (00:45). Those scenarios aren’t contrived.
“And this is the reality for over 200 million people” — Arjun
Macular degeneration, cataracts, and diabetic retinopathy affect hundreds of millions worldwide. Argus targets elderly people aging into vision loss who need answers without a phone to fumble with.
Hardware: Camera, YR, and Compute
Argus has three components: a camera module worn on glasses, a compute module, and a YR link between them. YR uses skin as a wireless transmission medium, charging and discharging capacitors to send data across the body.
“It uses 100 times less power than traditional Wi-Fi.” — Arjun
That efficiency gives the camera module a 10-hour battery life at just a few grams. The Build prototype weighs about 3 times more than the production target once the team scales manufacturing down.
The AI Stack: Edge Models and Cloud Fallback
On the edge, Argus runs sparse state-space models for voice recognition and convolutional neural networks for image processing. CNNs over transformers was a deliberate call: transformer edge AI accelerators don’t exist in usable form yet. A dedicated hardware accelerator handles each model type alongside a relatively weak central processor.
For complex tasks beyond simple scene queries, one button press connects to GPT-4o via Azure AI Foundry. A built-in SIM card handles cloud connectivity and pushes model updates from the team’s servers without requiring the user to enter a Wi-Fi password.
Privacy, Ethics, and Design Differentiation
YR transmission requires physical contact to intercept, so the camera-to-compute link is harder to sniff than Bluetooth or Wi-Fi. Facial recognition data stays on-device and never uploads to the cloud. Before any image reaches Azure, CNNs detect and blur faces.
Meta Ray-Bans require a cellular connection and a paired phone. Argus doesn’t. It has its own SIM, no Bluetooth setup, and haptic sensors at two angles to warn of walls and ledges automatically. The target user is an elderly person aging into vision loss who should not have to think about the technology at all.
Notable Quotes
And this is the reality for over 200 million people Arjun · ▶ 0:32
It uses 100 times less power than traditional Wi-Fi. Arjun · ▶ 4:34
But it’s smart, seamless and secure. Daniel · ▶ 9:38
Key Takeaways
- Argus uses body-area YR transmission at 100x lower power than Wi-Fi, enabling a 10-hour camera battery.
- A hybrid edge-cloud model runs sparse state-space and CNN models locally, escalating to GPT-4o for complex tasks.
- Facial recognition data never leaves the device; faces are blurred before any image reaches Azure.