Private AI search over support tickets, on the company's own server

Situation

Support agents researching a complex issue needed to find the past tickets that had already answered it. Searching the ticket database by hand was slow, and the ticket text could not be sent to an outside AI provider.

Solution

A search tool lets an agent type a question and see the past tickets closest in meaning. A chat assistant was also set up on the company's own server.

To choose a search method on evidence, Jeff Hunt built a version that ran each question through several methods at once. It showed the top tickets from each method side by side, so the support team could compare them on actual questions.

Design decisions

Because the data had to stay in-house, every model ran on private hardware, and the finished tools ran on the company's own server. The chat assistant was set up so the company could change AI models later without changing the application.

The engagement also produced a server specification for the company's administrator and written instructions for maintaining the services and installing newer models. Every software package was checked against the license terms in the company's contract.

Outcome

Support agents can find earlier answers to a complex issue by describing it in plain language. The ticket text stays in-house, and the company can change AI models later without rebuilding the tools.