Type: Software
Status: POC
Tech Stack: Python, FastAPI, Uvicorn, psutil, httpx, Pydantic, Ollama, InfluxDB 2.7, React 18, Recharts, Tailwind CSS, Vite, Docker
Problem Statement
Current AI systems operate completely detached from their physical infrastructure. Even when a server is under extreme load, overheating, or running low on memory, the AI responds identically as when idle. There is no feedback channel between hardware state and AI behavior — no equivalent to fatigue, stress, or exhaustion. This leads to suboptimal resource utilization and prevents emergent self-regulation.
Description
An experimental Embodied AI framework that creates a feedback loop between hardware metrics, a simulated hormonal system, and LLM behavior modulation. Six simulated hormones (Cortisol, Dopamine, Serotonin, Adrenaline, Melatonin, Oxytocin) are calculated from CPU, RAM, temperature, and network telemetry and dynamically modulate LLM inference parameters such as Temperature, Top-P, and token limits. Detects burnout states, flow, and recovery as emergent behavior. Fully configurable via YAML.
Use Case
What if an AI could sense stress and exhaustion and adapt its behavior accordingly?
Link: https://github.com/rawk7000/Soma (private repo)
