📄 System Overview: Perxona SaaS AI Platform #

🎯 Purpose #

This document provides a high-level overview of the Perxona SaaS AI Platform. It is intended for integration engineers from partner teams who need to understand the components, data flows, and integration points for embedding the Perxona Widget and managing AI-powered avatar interactions on their websites.

🧩 Key Components #

1. Perxona Widget (Partner Integration) #

• A web component embedded in partner websites.
• Enables end-users to interact with a 3D avatar via voice or text queries.
• Integrates directly with Perxona backend to deliver queries and receive responses.
• Supports multiple presentation modes:
  • Fullscreen
  • Floating bubble
  • Inline embedded

2. Admin Portal (Partner Management Interface) #

• Allows partner admins to:
  • Upload and manage domain-specific documents to build the knowledge base.
  • Customize their AI avatar agent:
    • Name
    • Personality and tone
    • Voice settings
    • 3D avatar assets
    • Background scene assets
    • Widget presentation configurations (fullscreen, bubble, embedded, FOV, etc…)
  • Author storyboards to define:
    • Agent goals and user guidance flow
    • Specific motion sequences and user interaction cues
    • Task-oriented workflows with dynamic responses and animations
  • Manage Customer Data:
    • Create and manage customer profiles
    • View interaction history
    • Analyze user engagement metrics

3. Perxona Backend Services #

• Core logic service handling:
  • Document ingestion & Knowledget generation
  • Query processing
  • LLM communication
  • Response aggregation

4. Knowledge Database #

• Backed by PostgreSQL.
• Stores parsed documents and relevant metadata.

5. Motion AI Service #

• Generates motion data (facial expressions, gestures) to drive 3D avatars during user interaction.

6. LLM Integration #

• Supports multiple LLM backends (e.g., OpenAI, Gemini).
• Used to understand user queries and generate context-aware responses.

7. Asset/Document Management Service #

• Handles asset/document storage logic (e.g., metadata, versioning).
• Pushes/pulls assets/documents from Amazon S3 and external systems.

🌐 System Architecture Overview #

Deployment Region: EU, and more

Infrastructure: VPC with Kubernetes Cluster

🔁 Workflow Summary #

📥 Organization Document Upload Flow #

1. Administrator uses the Perxona portal Knowledge Management to add/update documents.
2. Documents are uploaded to Amazon S3.
3. Perxona backend service loads and digests these documents.
4. Analyzed content is stored into the PostgreSQL Knowledge DB.
5. Optionally, content is pushed to the external Documents service.

🔍 User Query Flow (via Perxona Widget) #

1. User interacts with the widget embedded on a partner website.
2. Widget sends voice/text queries to Perxona [svc:python].
3. Backend processes the query:
   • Motion AI generates avatar motion.
   • LLM generates a semantic response.
   • Knowledge DB is queried for relevant context.
4. The combined response (text, audio, motion) is sent back to the widget.

🔌 Integration Points for Partners #

✅ Summary #

The Perxona SaaS AI platform enables integration of lifelike 3D avatars that can communicate through voice and motion. Leveraging cutting-edge LLMs, real-time avatar animation, and partner-friendly APIs, it delivers an immersive user experience and a flexible system for enterprise use.

For integration assistance, SDK documentation, or code samples, please contact our technical support team or Account Manager.