Built an internal AI knowledge assistant for a telecom BSS platform, plus the RAG ingestion and evaluation pipeline behind it, so technical consultants and pre-sales engineers get grounded answers from sprawling product documentation instead of digging through wikis and drives.
Context and stakes
At an enterprise software vendor, the technical consultants who configure and sell a complex telecom BSS and CPQ platform had to answer detailed product questions fast, with the documentation spread across wikis, drives, and object storage. A pre-sales beta launch put a hard deadline on making that knowledge usable through an assistant, with a customer demo as the acceptance gate.
Problem
Generic search and an ungrounded LLM were not good enough. Consultants need answers they can trust in front of a customer, and a wrong answer about product capability is worse than no answer. The documents were scattered and unstructured, retrieval accuracy was unmeasured, and there was no evaluation harness to tell whether a change made the assistant better or worse.