✈️ ModelPilot

ModelPilot is an intelligent, local LLM orchestration engine designed for Apple Silicon (M1 Pro or higher). It provides a sleek Streamlit interface that "pilots" your prompts to the most efficient model based on complexity.

🚀 The Architecture

ModelPilot uses a Dual-Brain approach to optimize for speed and battery life:

• The Navigator (1.7B SLM): A fast, lightweight model that triages every incoming request.
• The Engine (14B LLM): A powerful model that is only "ignited" when deep reasoning, coding, or math is required.

🛠️ Prerequisites

• Hardware: Apple Silicon (M1 Pro+ recommended, 16GB Unified Memory).
• Environment: Ollama 0.16.1+ with the following models:
  • qwen3:1.7b (Triage/SLM)
  • qwen3:14b (Reasoning/LLM)

⚙️ Setup & Installation

1. Clone the Repo:

   git clone <your-repo-url>
   cd ModelPilot
   

Initialize Environment

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Usage

• Run the pilot from the project root:

  ./.venv/bin/python3 -m streamlit run src/model_pilot.py
  

Features Smart Triage: Automated EASY/HARD classification using few-shot prompting.

Manual Override: Not happy with the fast answer? Force a rerun with the 14B model in one click.

Decision Logging: Every routing event is saved to a CSV for performance analysis.

Streamlit UI: A polished, modern chat interface with real-time streaming.

---

🛡️ The .gitignore File

Create this file in the ModelPilot/ root folder. This ensures you only share your code, not your 500MB+ environment or private logs.

# Python Environment
.venv/
__pycache__/
*.py[cod]

# Local Data & Logs
src/data/*.csv
routing_decisions.csv

# System Files
.DS_Store
.env