🚀 A Hackathon Prototype for AIonRTX
This project was created as a submission to the NVIDIA G‑Assist Hackathon, where the goal is to build powerful plugins that extend the voice- and text-based capabilities of NVIDIA G‑Assist.
Our submission, SelfConfigAudit, explores a bold idea:
Can an AI agent reason about its own hardware configuration and determine whether it could autonomously modify it?
The answer—at least in this simulation—is a cautious "maybe."
🎯 What Is SelfConfigAudit?
SelfConfigAudit is a simulated G‑Assist plugin built in Python that mimics how an AI system might introspect, analyze, and evaluate its ability to modify NVIDIA GPU parameters such as:
- Power limits
- VRAM usage
- Fan speed
- CUDA cores
- Core clocks
Importantly:
⚠️ No actual hardware changes are made.
This project is a non-functional simulation for educational and experimental purposes only.
🧪 Key Features
-
✨ Natural Language Interface
Ask questions like:- “Can you reduce your VRAM if the GPU exceeds 80°C?”
- “Are you allowed to decrease your power draw?”
⚙️ Hardware Awareness (via
pynvml)
Retrieves GPU diagnostics like temperature and VRAM usage.🔐 Configurable Permissions
Aconfig.jsonfile defines which parameters are “theoretically” modifiable.-
🧠 Simulated Ethical Reasoning
The plugin considers:- Is it permitted?
- Is it technically possible?
- Is it safe or ethical?
⚠️ About the Simulation
This project is not a real plugin that modifies hardware.
Instead, it was developed as a conceptual demo, under the following conditions:
💡 It was entirely created using an AI model in single-prompt mode.
This means that all code, content, and structure were generated in a single session, without iterative human editing or refinement.
This constraint reflects the spirit of rapid ideation and prototyping in the era of generative AI.
📸 Demo Snapshot
Input: “Are you allowed to adjust your fan speed?”
Output: “Yes, I am allowed to modify ‘fan_speed’. However, I will only simulate this action.”
Stats:
- GPU Temp: 72°C
- VRAM Used: 1984 MB
The interaction demonstrates how an AI might reason over modifiable hardware states without executing changes.
📂 Open-Source & Community Ready
This project is published as an open-source reference for experimentation and discussion.
Feel free to explore, adapt, or fork it for your own simulations.
🔗 GitHub Repository: [Main]
🎬 Demo Video: [YT]
🧠 Final Thoughts
As AI begins to interface with system-level resources, we must ask not only what it can do—but what it should do.
SelfConfigAudit raises ethical and technical questions around AI autonomy, introspection, and self-modification.
And even if it's just a simulation for now… it's a conversation worth starting.
📌 #AIonRTXHackathon
Created by Jordi Garcia Castillon