Prototype-stage embedded AI project · Taiwan

AI that lives beyond the screen.

EdgeTalk explores compact, voice-first AI interfaces for embedded devices—connecting microphones, displays, speakers and device actions to a cloud reasoning layer.

Explore the prototype See architecture
ESP32-S3Embedded controller
Voice-firstMicrophone → response
Modular I/OAudio · display · actions
Concept visualization of a compact EdgeTalk voice interface
Concept visualization — not a photograph of the current prototype.
01 / Product

A small-device interface for natural conversation.

The project focuses on a simple idea: let a physical device listen, reason in the cloud, and respond through the hardware around it.

◉

Listen

Capture voice through a USB microphone and manage interaction on an ESP32-S3.

✦

Reason

Send structured requests over HTTPS to a frontier model API. Claude is the initial target reasoning layer.

↗

Respond

Return spoken output, on-device visual feedback, or structured actions to connected hardware.

02 / Prototype

Built around real embedded hardware.

Current work is centered on an ESP32-S3 development platform with audio I/O, a compact TFT display, networking and device-state control.

Hardware foundation

Prototype stack

  • ESP32-S3 development board
  • USB microphone input
  • 2.2-inch SPI TFT display
  • I²S audio amplifier + speaker
  • Wi-Fi / HTTPS connectivity
  • FreeRTOS-based device logic

Development status

✓
Embedded I/O foundation

Audio, display, networking and device-side control are the core prototype layer.

↻
Cloud reasoning integration

Claude API integration is planned as the next development milestone.

→
Productization

Enclosure, onboarding, reliability and repeatable deployment are future milestones.

03 / Architecture

From a voice input to a physical response.

A deliberately modular pipeline keeps the embedded layer lightweight while allowing the reasoning layer to evolve.

1⌁User voiceNatural input
→
2◉MicrophoneAudio capture
→
3▣ESP32-S3Device control
→
4⇄Speech / promptTranscription + context
→
5✦Claude APITarget reasoning layer
→
6↗OutputsVoice · UI · actions

Integration note: Claude integration is planned. Speech recognition and speech synthesis are separate services in the proposed architecture.

04 / Use cases

Where voice can be more natural than another app.

⌂

Smart-device interfaces

Natural-language control and feedback for small connected devices and appliances.

⌘

Educational prototyping

A hands-on bridge between embedded systems, AI APIs and human-device interaction.

◫

Custom AI hardware

A reusable reference architecture for voice-first prototypes with displays, audio and device actions.

Why Claude

A reasoning layer for interactions that go beyond chat.

The project is evaluating Claude as the first cloud reasoning layer for contextual conversation and structured device responses. Program support and API credits would be used to move from embedded I/O experiments toward a repeatable end-to-end prototype.

Context-aware conversationStructured responsesTool / device-action workflowsRapid prototype iteration
05 / About

Independent, early-stage, built in Taiwan.

EdgeTalk Labs is an independent prototype project in Taiwan, exploring voice interfaces for embedded devices.

The project is at the personal-prototype stage, with a goal of developing a repeatable reference platform for makers and small device teams. EdgeTalk Labs is a working project name and is not currently an incorporated company.

Next step

Turn the prototype into an end-to-end AI device.

The next milestone is a voice-to-response demonstration on hardware, followed by measured latency, reliability, and cost evaluation.

Back to top ↑