PrismaAI: Multi-Agent Intelligent Copilot

PrismaAI is an advanced vehicle assistance system developed with langchain and langgraph. It leverages a multi-agent architecture to manage distinct knowledge domains—navigation, driver profiling, and technical diagnostics—ensuring long-term persistence and human control over critical decisions.

🏗️ System Architecture

The PrismaAI state graph is designed following the Supervisor/Router pattern, where a central node orchestrates the activation of specialized agents based on user intent.

⚠️ This repository is an early architectural sketch of a larger Intelligent On-Board System.

The current implementation uses Jupyter Notebooks as an experimentation and prototyping environment to validate core ideas, agent orchestration flows, and RAG strategies before moving to a production-grade codebase.

Graph Components:

• Decision_Router: The entry point that analyzes the request and directs the flow to the appropriate specialist agent.
• Specialized Agents:
  • NavigationAI: Manages routes, traffic data, and travel preferences.
  • PrismaCopilot: Focuses on personalized interaction and maintaining the driver's identity.
  • LogCarManualAI: Expert in technical manuals and vehicle maintenance history.
• Human_Feedback: A Human-in-the-Loop interruption node that allows for human validation before task completion or flow redirection by the router.

---

💾 Persistence & Long-Term Memory

A core pillar of this project is the use of isolated, persistent memories. This allows the system to remember user preferences and historical vehicle issues across different sessions/chats. In this project its stored during execution time (RAM), but is adaptable to use SQLITESaver, for database use.

Agent	Memory Scope	Description
NavigationAI	routes_preferences	Stores frequent destinations and route settings (e.g., "avoid tolls").
PrismaCopilot	user_profile	Maintains profile data, hobbies, and personalization.
LogCarManualAI	historic_car_problems	Records diagnostic history and vehicle health logs.

---

🛠️ Technologies Used

• Python 3.11+
• LangGraph: For state graph orchestration and persistence.
• LangChain: Framework for LLM integration and tool calling.
• Human-in-the-Loop: Implementation of breakpoints for real-time feedback.

---

🚀 Getting Started

1. Clone the repository:

   git clone [github_project]

2. Install dependencies:

   pip install -r requirements.txt

3. Configure environment variables: Create a .env file and add your API keys:

   OPENAI_API_KEY=your_key_here
   GOOGLE_DIRECTIONS_API=your_api_here

4. Run the system:

   • Open main.ipynb
   • Execute each cell

---