AI Engine Architecture

Overview

Vartovii includes a built-in AI assistant powered by Google Vertex AI (Gemini).

System Architecture

backend/ai/
├── config.py           # Model & region configuration
├── client.py           # GeminiClient singleton
├── services/
│   ├── chat_service.py      # Chat assistant (RAG + Tools)
│   ├── report_service.py    # PDF Report generation
│   └── sentiment_service.py # Review analysis
└── tools/
    ├── crypto_tools.py      # Blockchain data
    └── search_tools.py      # Google Grounding

Models Strategy

Task	Model	Reason
Chat Assistant	gemini-3-flash-preview	Low latency, high throughput (with 2.5 fallback)
Smart Reports	gemini-3.1-pro-preview	Complex reasoning, large context (with 2.5 fallback)
Sentiment	gemini-3-flash-preview	Cost efficiency for bulk processing
Visuals	Imagen 3	Generating social media assets

API Endpoints

AI Chat

POST /api/ai/chat

• Payload: {"message": "Analyze BMW", "language": "ua"}
• Features: Function Calling, RAG (Vector Search), Context Retention.

Smart Reports

POST /api/reports/generate

• Payload: {"company_slug": "bmw"}
• Output: PDF Download URL.

RAG & Knowledge Base

The AI Engine is grounded in real-time data from:

1. PostgreSQL: Trust Scores, Financials, Reviews.
2. Vector Store: Indexed internal documentation (Knowledge Base).
3. Google Search: Live web grounding for recent events.

Configuration

Managed via backend/ai/config.py and environment variables:

• PROJECT_ID: sentry-analytics-prod
• REGION: europe-west4 (for data residency)