Arvel

Inspiration

You sit down to study, get stuck on a concept, and reach for your phone to look it up. A notification or feed is right there too. What was meant to be a quick question can turn into doomscrolling.

I wanted a desk buddy that gives me somewhere else to ask for help, so I can leave my phone aside while studying. That is the problem Arvel is built around.

What it does

Arvel is a small, voice-enabled study companion. Ask it to explain an unfamiliar concept, then follow up on the part you still do not understand. You can speak or type on its touchscreen, and read or hear the reply.

For example: "Explain fractions with a simple example." The goal is to help with the question in front of you without opening a social feed along the way. Arvel does not block apps or lock your phone, and its answers should be checked against your course material.

An agent running on a small ESP32

The Rust agent runtime runs directly on the ESP32-S3, alongside the touchscreen, microphone and speaker. It keeps the conversation, manages the model and tool loop, and dispatches device tools locally. This is an agent running inside the desk buddy, not on a laptop hidden behind it.

Model inference still happens in the cloud; the language model does not run on the ESP32. Voice input and spoken replies make the device usable without typing every question, and microphone activation stays under physical user control. No computer needs to stay running for Arvel to work.

Built With

• c
• c++
• esp-idf
• esp32-s3
• espeak-ng
• freertos
• lvgl
• ollama
• rust
• sveltekit
• tailwind
• typescript
• vite