The user’s question is sent via WebSocket (WS) to the backend
server, which triggers the invocation of a processing chain. The specific chain that is
activated will depend on the selected focus mode. In this scenario, we will utilize the
“Enterprise Search” focus mode
Once the chain is invoked, the user’s message is
forwarded to another processing chain. This chain assesses the chat history alongside
the current question to determine if there is a need to source information from the
Enterprise data repo.
If a query is generated, it is sent to a search engine, to retrieve
relevant information from the Enterprise data source. A similarity search is then conducted
to identify the most pertinent sources that can effectively address the user’s query.
After the relevant sources have been identified, they are
forwarded to the response generator chain. This chain synthesizes the chat history, the
generated query, and the gathered sources to formulate a coherent response. The
final answer is then streamed back to the user interface (UI) for the user to view.
Once the chain is invoked, the user’s message is
forwarded to another processing chain. This chain assesses the chat history alongside
the current question to determine if there is a need to source information from the
Enterprise data repo. If it concludes that sources are necessary,