Secure RAG Knowledge Search for a Utility Company

A utility company was drowning in its own documentation. Finding the basic information needed for a critical procedure could take an employee up to six hours.

The knowledge existed, but it was buried across mountains of technical documents with no fast, reliable way in. I built a secure, local, multimodal RAG search engine tailored to their infrastructure data, using knowledge graphs to map how information connects and return accurate, sourced answers. Research that had taken around six hours dropped to under three minutes, and every answer was fully auditable.

The challenge

In a safety-critical, regulated environment, slow information is not an inconvenience. It is a risk.

• Critical procedures depended on information scattered across vast technical documentation.
• Employees spent up to six hours locating what they needed, delaying decisions that mattered.
• Answers could not just be fast. They had to be accurate and traceable, because the work has compliance and safety consequences.

A keyword search over a document dump would not solve this. The company needed answers it could trust and verify, drawn from connected technical knowledge, without sensitive infrastructure data ever leaving its control.

The approach

I led with the two non-negotiables for this setting, security and auditability, and built the retrieval system around them.

I: Implementation Planning

I designed the system to run locally on the company’s own infrastructure, so sensitive data never left the building. It had to handle multimodal documents and trace every answer back to its source. To return accurate answers rather than keyword matches, I planned a knowledge-graph approach that maps the connections across the documentation.

M: Migration & Execution

I built a secure, local, multimodal RAG search engine tailored to their infrastructure data, with knowledge graphs mapping how information relates so the system could deliver accurate, sourced answers instantly. Every result pointed back to the document it came from.

The results

Six hours to under three minutes. Research time for critical information collapsed from roughly six hours to less than three minutes.

Auditable by design. Every answer traced back to its source, so the team could verify what the system told them, as a regulated environment requires.

Faster critical decisions. With trustworthy answers available instantly, decisions that had waited on document research moved ahead.

Why this matters

In regulated industries, a knowledge assistant is only useful if you can trust it and prove where its answers came from. A fast tool that cannot show its sources is unusable for safety-critical work.

Running retrieval locally and making every answer auditable is what turns RAG from a demo into something a utility can actually rely on. The speed is the headline. The traceability is what makes the speed safe to use.