Enterprise AI Platform with RAG & Zero-Trust
Client: Enterprise AI Client
The Problem
Traditional enterprise AI platforms require exposing databases to the cloud or installing invasive on-premise software. The client needed RAG over PostgreSQL, MySQL and MSSQL with text-to-SQL, document chat and multi-LLM routing — all with zero-trust architecture.
Security challenges included prompt injection, PII leakage and unauthorized SQL access.
The Technical Solution
Chourio DEV designed a 5-service platform:
- AI Engine (FastAPI + Haystack 2.x + Qdrant): RAG-SQL, RAG-Docs and Chat pipelines with multi-LLM routing (OpenAI, Anthropic, Google, Cohere)
- 3-layer guardrails: anti-injection, SQL scope validation and PII sanitization
- SaaS Backend (Express + Prisma): 35+ modules — auth, onboarding, agent flows, knowledge base
- On-premise Gateway Agent (Node.js): zero-trust read-only bridge with job polling — outbound HTTPS only
- Gateway Provisioner (Go): Docker orchestration with dormancy/wake supervisor
Business Impact
- Natural language queries on on-premise data without exposing credentials
- Multi-LLM routing with fallback chains for high availability
- Text-to-SQL pipeline with semantic schema linking
- Platform ready for international multi-tenant deployment