🚀 AI Event Intelligence Engine

A Retrieval-Augmented Generation (RAG)-inspired backend system for contextual event risk analysis using vector similarity search.

---

📌 Overview

AI Event Intelligence Engine is a backend system that performs contextual risk assessment on operational and security events.

The system:

• Converts events into vector embeddings
• Stores them in PostgreSQL with pgvector
• Retrieves similar historical incidents using vector similarity
• Aggregates contextual signals
• Produces structured risk scoring with confidence metrics

This project demonstrates practical AI engineering principles including vector databases, contextual retrieval, and structured decision modeling.

---

🧠 Architecture

Client Request ↓ FastAPI Backend ↓ Embedding Generator ↓ Supabase PostgreSQL + pgvector ↓ Vector Similarity Retrieval ↓ Context Aggregation ↓ Risk Scoring Engine ↓ Structured JSON Output

---

🛠️ Tech Stack

• FastAPI
• PostgreSQL (Supabase)
• pgvector
• SQLAlchemy
• Python
• REST API

---

📡 API Endpoints

GET /

Health check endpoint

---

GET /db-test

Tests database connectivity with Supabase.

---

POST /events

Stores a new event with embedding and base risk classification.

Example:

{
  "description": "Fire alarm triggered in Block A"
}
POST /search
Retrieves top similar historical events using vector similarity.

POST /analyze
Performs contextual risk analysis using retrieved similar events.

Example Response:

{
  "input_event": "fire alarm trigger",
  "risk_assessment": {
    "risk_level": "LOW",
    "confidence_score": 0.11,
    "similar_events_considered": 3,
    "average_distance": 15.72
  },
  "retrieved_context": [...]
}
🎯 Key Features
Vector-based similarity search

Context-aware risk scoring

Confidence calculation

Structured decision output

Cloud database integration

Modular backend architecture

🧪 Development Mode
Current implementation supports deterministic/mock embeddings for development and architectural validation.

In production environments, semantic embeddings (OpenAI or local LLMs) can be integrated.

📈 Future Improvements
LLM-powered reasoning layer

Time-decay weighted scoring

Real-time streaming ingestion

Multi-tenant architecture

Dashboard visualization

Async background embedding pipeline

🎓 Learning Outcomes
This project demonstrates:

Applied vector database implementation

Retrieval-Augmented architecture design

AI system decision modeling

Production-style backend engineering

Cloud database connectivity & pooling