LangGraph and Multi-Agent Architecture in Prisma AI: The Power of Intelligent Coordination

Introduction

In Prisma AI's architecture, instead of using a single language model to handle all requests, we implement a Multi-Agent architecture powered by LangGraph. This system operates as a State Machine, where specialized AI Agents coordinate seamlessly through a Shared Memory called State to complete complex tasks with high accuracy and aesthetic quality.

Below is how the trio "Planner - Narrator - Designer" collaborates to transform raw data into professional presentations and reports.

1. Strategic Planner: The "Architect" of Planning

Every process begins with the Strategic Planner, an agent that acts as a strategic consulting expert.

Deep Analysis

This Agent scans through all source documents and user queries to identify:

• Target audience
• Appropriate tone
• Communication objectives

Framework Design

The Planner builds a logical Outline with strict rules:

• Each heading must be a complete sentence
• Concise and no more than 12 words
• Ensuring professionalism and coherence

┌─────────────────────────────────────┐
│         STRATEGIC PLANNER           │
├─────────────────────────────────────┤
│  Input: Source docs + Query         │
│                                     │
│  ┌─────────────────────────────┐    │
│  │  1. Analyze audience        │    │
│  │  2. Determine tone          │    │
│  │  3. Build Outline           │    │
│  └─────────────────────────────┘    │
│                                     │
│  Output: Logical Outline            │
└─────────────────────────────────────┘

2. Narrative Story Agent: The "Storyteller" of Content

Once the framework is ready, the Narrative Story Agent (Narrator) takes over to breathe life into the content.

Knowledge Extraction

Using Long Context LLM models, this agent "reads" deeply into documents to write key points for each section:

• Extract important information
• Synthesize data from multiple sources
• Create coherent and logical content

Logical Thinking

Instead of just listing raw data, the Narrator focuses on answering the question "Why?" to:

• Prove and defend the core message
• Build persuasive arguments
• Connect ideas logically

3. Visual Reasoning & Layout Agent: The "Artist" of Visuals

When content is ready, the aesthetic-focused agents (Designer) join to optimize the visual experience.

Visual Reasoning Agent

AI analyzes content to make intelligent decisions:

Case 1: Processes or hierarchical structures

• Automatically generate Mermaid code to draw diagrams
• Visually represent workflows
• Show relationships between components

Case 2: Abstract concepts

• Compose prompts for DALL-E 3 to create unique illustrations
• Design images appropriate to the context
• Enhance content visualization

Layout Design Agent

This agent determines the optimal page layout based on the presence of visual assets:

Perfect Coordination Mechanism via LangGraph

This coordination is not random but tightly controlled through LangGraph's data flows.

State Machine Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   PLANNER    │────▶│   NARRATOR   │────▶│   DESIGNER   │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       ▼                    ▼                    ▼
┌─────────────────────────────────────────────────────────┐
│              PRESENTATION STATE (Shared Memory)         │
├─────────────────────────────────────────────────────────┤
│  • Outline          • Content           • Visuals       │
│  • Target Audience  • Key Messages      • Layouts       │
│  • Tone             • Arguments         • PPTX Output   │
└─────────────────────────────────────────────────────────┘

Controlled Data Flow

Data flows in a controlled manner between steps, allowing agents to:

• Read from the shared PresentationState
• Write results to State after each processing step
• Synchronize information between Agents

This ensures the entire system remains consistent from the planning stage to the final PPTX file export.

Conclusion

The Multi-Agent architecture with LangGraph in Prisma AI demonstrates the power of intelligent coordination:

• Planner ensures clear strategy and structure
• Narrator creates content with depth and logic
• Designer delivers professional visual experience

This combination not only produces high-quality outputs but also ensures consistency and scalability for complex tasks in the future.

Want to experience the power of Multi-Agent architecture in Prisma AI? Contact us for consultation and product demo.