AI Knowledge & Search System
Making a 13,000-Agent Training Library Searchable in Plain Language
INDUSTRY
Real Estate
capability
AI & Intelligent Tools
The organisation supports a network of more than 13,000 agents with training material, scripts, templates, and recorded sessions accumulated over years. That library lived across four places: a content management system, a production database, Google Drive, and video.
Every resource an agent needed already existed. The problem was reaching it.
The Challenge
Content was spread across systems with no shared index. Search returned document titles rather than answers, which meant an agent had to already know what a resource was called in order to find it. For a network of that size, with new material added continuously, the library grew faster than anyone's ability to navigate it.
Any solution had to work with the content platform already in place. Replacing it was not an option, and neither was asking 13,000 people to change how they worked.
What We Built
A conversational search application layered on top of the organisation's existing content platform.
Automated ingestion. An hourly process scans for new and updated material, extracts text from Docs, Sheets, PDFs, video and the database, classifies each item by resource type, and refreshes the index. No manual upload step.
Natural-language search. Queries pass through intent detection, retrieve relevant material from an embedding index, and return refined answers alongside the source resource.
Clarifying questions. Where a query is ambiguous, the system asks rather than guessing.
Conversational context. Follow-up questions build on the previous answer.
Single sign-on with one-time password verification, matching existing user accounts.
Feedback routing into the team's working channels, with a separate administrator portal.
The Outcome
Built, deployed, and serving the organisation's full agent network. Four content sources sit behind one search interface, and the index keeps itself current without anyone maintaining it.
13,000+
agents supported
4 systems
unified in one search
Hourly
automatic index refresh
The Challenge
Content was spread across systems with no shared index. Search returned document titles rather than answers, which meant an agent had to already know what a resource was called in order to find it. For a network of that size, with new material added continuously, the library grew faster than anyone's ability to navigate it.
Any solution had to work with the content platform already in place. Replacing it was not an option, and neither was asking 13,000 people to change how they worked.
What We Built
A conversational search application layered on top of the organisation's existing content platform.
Automated ingestion. An hourly process scans for new and updated material, extracts text from Docs, Sheets, PDFs, video and the database, classifies each item by resource type, and refreshes the index. No manual upload step.
Natural-language search. Queries pass through intent detection, retrieve relevant material from an embedding index, and return refined answers alongside the source resource.
Clarifying questions. Where a query is ambiguous, the system asks rather than guessing.
Conversational context. Follow-up questions build on the previous answer.
Single sign-on with one-time password verification, matching existing user accounts.
Feedback routing into the team's working channels, with a separate administrator portal.
The Outcome
Built, deployed, and serving the organisation's full agent network. Four content sources sit behind one search interface, and the index keeps itself current without anyone maintaining it.