Uber's engineering teams face a constant stream of routine support questions. Platform teams that build tools for other developers, such as deployment, data processing and ML workflows, typically assign an on-call engineer to answer questions that arrive through Slack channels. Even simple requests can take several rounds of back-and-forth before the right documentation is found or the steps are explained. Uber reports that its internal support channels receive approximately 45K questions each month, consuming substantial engineering time that could be spent on more valuable work.
Much of the information needed to answer these questions already exists, but it is scattered across Engwiki, Uber's internal engineering wiki, its internal Stack Overflow and other documents. Locating the right material is hard enough that asking a colleague is often easier. Genie is designed to make this existing knowledge available directly in the Slack channels where developers already seek help.
Genie is Uber's internal AI assistant for engineering support. It finds relevant information in Uber's documentation, uses a language model to explain that information, and lets engineers continue the conversation or escalate to a human. A general-purpose language model can write clear explanations, but it lacks knowledge of Uber's private platforms, internal procedures and documentation. Genie closes that gap by combining retrieval with generation.
The system prepares knowledge, retrieves relevant material, generates answers and checks their correctness, with a path to a human expert when it cannot respond. Uber also collects user feedback and evaluates Genie's performance to improve it over time.
Source: ByteByteGo · Summarized by HeadlinesBriefing