Institutional knowledge is useful only if you can find it.
A service archive can hold years of valuable answers while still being frustrating to use. The case title may describe one issue, the useful detail may be in a technician's work note, and a search result can lose the chronology that explains what actually happened.
For Illumia, I built an internal tool around that retrieval problem. The objective was a dependable way to find prior work and understand its context, with AI as another interface to the same evidence rather than a separate source of invented answers.
Search the details. Return the whole story.
The local application imports the service archive into PostgreSQL while preserving original records and the relationship between parent cases and child work entries. The browser interface supports literal and keyword search, combined filters, customer and component context, case and work-note dates, technicians, and recorded time.
Results are grouped by parent case. Matching child records are highlighted within the case history, with anchored navigation and pagination that preserve context. A Windows launcher and additive migrations support repeatable use without discarding the local database or configuration on an update.
Give the assistant useful tools, not unrestricted access.
The Claude Desktop connection exposes six bounded, read-only tools for archive information, filter lookup, search, case retrieval, work-record retrieval, and exact aggregates. The browser and the assistant use the same retrieval service, so a change in interface does not create a different version of the evidence.
Counts and totals come from database queries rather than model guesses. The assistant does not receive arbitrary SQL, shell, or filesystem-write access. Retrieval is local, while selected evidence can be sent to online Claude when approved for the organization; the full AI workflow is not described as offline.
An implemented tool with a deliberate rollout boundary.
The application, packaging, and MCP startup have been tested with synthetic records. The release notes distinguish that implementation evidence from full-scale archive relevance checks and acceptance on managed workplace devices. Real company data and credentials are not bundled into the code or test fixtures.
The next step is validating retrieval against the actual archive and the organization's operating requirements. The aim is practical: help someone find the relevant earlier case, inspect the supporting work, and answer a question with a traceable source.
About this case study +
Illumia repository README, implementation status, search and mapping documentation, local-import design, and MCP retrieval boundaries. Validation described here uses synthetic records; workplace rollout is a separate acceptance step.
Researched and updated October 3, 2026.