Technology

Stop-Hack-Go has two sides: hands-on learning with Pico and electronics, and a classroom AI chatbot on a closed network. Everything stays on your network; no cloud required.

Technology stack

Learner hardware

Raspberry Pi Pico with LEDs, resistors, cables, and a multimeter. Students build circuits, write MicroPython code, and explore cybersecurity and coding concepts on real hardware.

Chatbot server

Raspberry Pi 5 (4GB or 8GB RAM) with the AI HAT+ 2 (Hailo-10H NPU) for fast, local LLM inference. Students interact with the chatbot via browser; all compute stays on the Pi in a closed network.

Backend

Flask serves the web app; Gunicorn is used for production. The stack is Python 3, with configuration and environment handled via config modules and environment variables.

AI / inference

Hailo-Ollama provides local LLM inference (e.g. qwen2.5) using the Hailo NPU. All model runs happen on-device—no data is sent to the cloud.

RAG

ChromaDB stores document embeddings. Teachers and admins upload project docs; the app indexes them and retrieves relevant chunks to add to the AI prompt, so chatbot answers are grounded in your materials.

Deployment

The application runs in Docker for a consistent environment. SQLite is used for app data (e.g. teams, sessions) where needed. Once the Pi is set up, the chatbot is accessed at a single URL on your closed network.

Architecture at a glance

The web server queries the vector store for context and sends the prompt plus context to Ollama; the NPU runs inference and the response is streamed back to the student.