Hybrid RAG Question Answering System

Democracy’s cornerstone is parliamentary debate, but having to wade through red tape, hours of recordings or pages of transcripts to find answers means it is nearly impossible for anyone to truly understand what’s being discussed.

As part of the ongoing development of The World Avatar™, our German subsidiary CMPG has recently deployed a hybrid RAG approach to the BMUKN (Federal Ministry for the Environment, Nature Conservation and Nuclear Safety) in Germany.

Debates in the German federal parliament are transcribed and made freely available to the general public. However, these are difficult for the average citizen to parse, and to explore. For example, if someone wanted to know ‘which political parties frequently argue in favour of share-based pensions?’ this would be incredibly time consuming to discover from exploring transcripts.

In partnership with BMUKN and the EU, a hybrid RAG solution involving semantic knowledge graphs and an LLM was deployed which translated plain text queries into semantic queries, augmented with contextual information. Data was then retrieved from a knowledge graph and returned to the user.

This allows for complex queries, and for better insights into the political process.
 
The tool, and some example queries, can be accessed at https://qa.theworldavatar.io/.