Arduino UNO Q Customer Service Agent
Two Arduino UNO Q boards running a customer service desk with on-device language models
This project runs two small language models on two Arduino UNO Q boards. One board writes customer support messages and the other classifies each message and replies with the matching answer from a knowledge base, all on the local network.
The Arduino UNO Q combines a Qualcomm Dragonwing QRB2210 running Debian Linux with an STM32U585 microcontroller. Each board runs Gemma 3 1B on the Linux side.
The customer board picks one of six problem areas and its model writes the message. The agent board classifies the message into one of six intents and replies with that intent's knowledge base entry word for word, so the model never writes a fact. A quarter of replies answer the wrong question on purpose, so the demo shows a bad experience as well as a good one. Over a 33 ticket run the customer reported 23 good outcomes against 7 deliberate mistakes.
The boards find each other on the network automatically and each serves a live view of the conversation in the browser. The repository includes the application and links to 3D printable cases for the boards.
Measured results.
- Tickets in test run
- 33
- Good outcomes reported
- 23
- Deliberate mistakes
- 7
- Model per board
- Gemma 3 1B
Let us build your future
Bring us a plan or just an idea. We will arrange a discovery meeting with you to go through it.