Deflecting 45% of IT Support Tickets with a VPC-Isolated RAG Agent

The Challenge

A leading US-based FinTech platform was struggling with a bloated IT Helpdesk. Their 500+ employees were constantly generating support tickets for routine queries regarding internal HR policies, API documentation, and compliance protocols. The information was scattered across dozens of SharePoint folders and Confluence pages, making it impossible for staff to find answers quickly. The client required an AI Assistant, but due to strict financial regulations, no proprietary data could be sent over public internet endpoints (like standard ChatGPT).

The Acadify Solution Architecture

Acadify's engineering team deployed a highly secure Retrieval-Augmented Generation (RAG) agent completely isolated within the client's AWS Virtual Private Cloud (VPC).

• Secure Embedding Pipeline: We built automated Python ETL scripts to ingest documents from SharePoint, chunk them semantically, and generate embeddings using open-source models running locally on EC2.
• Private Vector Database: Embeddings were stored in a managed AWS RDS PostgreSQL instance with the pgvector extension, ensuring data never left the VPC perimeter.
• Private LLM Execution: We utilized AWS Bedrock (Anthropic Claude 3 Haiku) for generation, which guarantees zero-data retention by the foundational model provider.
• Slack Integration: The agent was wrapped in a Slack bot, allowing employees to ask complex questions directly in their workflow.

The Impact

Within the first 30 days of production deployment, the RAG agent successfully answered over 3,000 employee queries, resulting in a 45% drop in Tier-1 IT support tickets. The IT team was able to reassign two full-time helpdesk engineers to core platform development. Furthermore, the architecture passed the client's rigorous internal InfoSec audit without a single red flag.