AI Knowledge Management System

Description

An engineering consultancy of about 300 people. Our institutional knowledge lives in project folders named after clients, in the heads of six senior engineers, and in a Confluence nobody has updated since 2023. We want search that works across the project archive and returns the specific calculation, drawing or method note somebody needs, with the project it came from and the year. The year matters more than people expect — a method from a 2011 project may be against current standards, and the system must show its age rather than presenting everything as equally current. Client confidentiality is absolute. Project material must only surface to people staffed on that project or its successors, and we will test that before anything goes live.

• Index the project archive with project, date and document-type metadata preserved
• Enforce project-level access control in retrieval, tested with explicit negative cases
• Surface document age prominently and flag material that predates current standards
• Handle the mixed formats in the archive, including CAD exports and scanned calculation sheets
• Provide a feedback path for engineers to mark a result as superseded

• Search across the project archive with metadata-aware results
• Access control enforced at retrieval, with a passing negative-case suite
• Age and standards flagging
• Superseded-content feedback workflow