AI document intelligence for a national energy operator

Certificates and compliance documents scattered across multiple email inboxes with no systematic organization. Staff members spent 10 to 15 hours each week sorting and filing documents by hand.

Retrieving a specific document during a regulatory inspection required 5 to 15 minutes of searching through email attachments, creating real compliance risk when inspectors needed answers in seconds.

No tracking system for certification expiration dates meant the operations team discovered expired documents only when a regulator flagged them, putting equipment operation and business continuity at risk.

Built an automated email ingestion system that monitors two dedicated inboxes, extracts attachments, and passes them through an AI classification pipeline. The system identifies document type, equipment unit, and relevant metadata without human intervention.

Created a vector database using pgvector that stores document embeddings alongside traditional metadata. This enables both structured queries (find all certificates for equipment unit G07) and natural language search (show me documents expiring in the next 30 days).

Deployed a chatbot interface that gives field staff instant access to any document through conversational queries. The system returns direct download links with previews, eliminating the need to dig through folder structures.