How johnnysgithubaccount/smart-home-virtual-assistant works
Project Type and Purpose
• From the README and file tree, Marvin is a voice-activated, multi-agent smart-home assistant. It’s a full-stack, graph-based AI system that listens for a wake word, classifies intent via SLMs, routes to chat or device-control agents, and persists context in a graph database. It also includes a minimal web backend for optional UI or API integrations and low-level hardware control for ESP32 devices.
File Structure
Top-level files/directories:
• Home Assistant/ – main Python package
– components/ – core modules (LLM wrappers, scheduler, long-term memory, knowledge-graph, logging, tools)
• hardware/ – device-control adapters & ESP32 C programs
• voice/ – speech-to-text, text-to-speech, wake-word detection
• llm.py, knowledge_graph.py, scheduler.py, nodes.py, tools.py, utils.py
– init_graph/ – setup scripts for LangGraph workflows (autonomous.py, instance.py)
– main.py – entry point
– deploy.py – deployment orchestration
– configs.json – external service configs
– web/ – lightweight backend (backend.py) for UI or REST APIs
– test/ – unit tests
• README.md, requirements.txt, .gitignore, etc.
README for Architectural Insights
• Components: SLMs served by Ollama, LangGraph for orchestration, Neo4j for memory, Firebase for real-time device control, HuggingFace wake-word engine, ESP32 firmware.