Nexus AI-Core Engine

A production-ready Modular RAG (Retrieval-Augmented Generation) system built with NestJS and PostgreSQL.

Core Capabilities

• Vector-First Retrieval: Native integration with pgvector for scalable semantic search.
• Reliability Layer: Built-in confidence scoring engine to validate AI outputs and prevent hallucinations.
• Operator Handover Protocol: Standardized flag system for escalating low-confidence queries to human support (CRM/Chatwoot compatible).
• Background Processing: Decoupled document indexing and embedding generation via BullMQ/Redis.
• Knowledge Ingestion: Advanced pipeline for processing unstructured data (PDF, DOCX, TXT).

Tech Stack

• Framework: NestJS & TypeScript
• Persistence: PostgreSQL + pgvector
• Workflow: BullMQ (Redis)
• Intelligence: OpenAI / Anthropic / Custom LLM Bridges

Architecture

The system is designed as a standalone "Core Engine" that communicates with integration bridges. It focuses on the logic of retrieval, scoring, and response generation, ensuring high modularity and ease of maintenance.