Expand description
Plato Nervous System — Room-Specific Model Distillation
Each room starts with big LLM calls. As tiles accumulate, tiny distilled models take over. Eventually each room trains its own LoRA — the irreducible intelligence that can’t be pure algorithm.
Signal chain: Sensor → Deadband → Nano → LoRA → Fleet → Cloud Each layer resolves what it can and passes the rest up.
Modules§
- ollama
- Ollama Integration for PLATO Nervous System
Structs§
- Deadband
Filter - Distillation
Config - Distillation
Record - Distillation
Stats - Jepa
Nano - The JEPA (Joint Embedding Predictive Architecture) nano-model This is the irreducible core — the room’s self-model. It doesn’t process language. It processes embeddings. Its job: predict the NEXT room state from the current state. When predictions diverge from reality, that’s anomaly detection that no threshold or rule could ever catch.
- Jepa
Nano Config - Model
Config - Nano
Model - Nervous
System Stats - Resolution
Distribution - Room
Nervous System - Room
State Vector - Rule
- Sensor
Reading - Tile
- Tile
Example