Inspiration
We got inspiration from the Tamagotchi which is a pocket-sized digital pet simulation toy. It was popular in the late 90s and we wanted to bring it back to life with a layer of AI to make it better. We also tried to be really sustaintable. Coming from Sheffield, we didnt have access to any 3d printers or industry tools, instead we used cardboard boxes that they used to give out crisps during hacknotts, the plates and other cutlery as tools to make our final product.
What it does
Iki combines conversation with physical interaction. You can talk to it, pet it, make it play music and use it to as a digital assistant to help you do tasks.
How we built it
We combined a local language model running through Ollama with an Arduino, a Grove RGB LCD, two ear servos, a joystick, a touch sensor, a motion sensor, a potentiometer and a buzzer. We also had used a raspberry pi and EPS32-C6.
A phone web app provides the microphone and speaker, while ElevenLabs generates spoken replies. A Python serial bridge connects the Arduino to our backend.
Iki is powered by hermes harness running on the Qwen 3.5 9B model running locally. It has access to the web, can help you remember tasks, and even do some of them. You can highly customize the way he speaks and his looks.
For the enclosure, we worked with recycled cardboard and available materials, giving Iki a handmade retro identity.
Challenges we ran into
Getting the phone, AI backend running split on Raspberry Pi and laptop, and Arduino to communicate reliably was a major integration challenge.
Building an appealing enclosure without access to a 3D printer pushed us to be resourceful.
Accomplishments that we're proud of
We brought conversation, memory, expressive animations and physical interaction together in a working prototype. We have no idea how we managed to get everything to interact well together without breaking anything. We are also proud by how we manage to get everything built despite not having much. We had no 3d printers or building material but used cardboard boxes from the lunch that we were provided with.
What we learned
We also gained practical experience connecting local AI, browser audio and embedded hardware.
Built With
- api
- arduino
- c++
- elevenlabs
- esp
- hermes
- mcp
- node.js
- ollama
- python
- raspberry-pi
- typescript
Log in or sign up for Devpost to join the conversation.