Who is this for?

Policy makers who need to see how ideas move through our local government Advocates who want to hold politicians accountable to their previous statements People who want to quickly get informed on a particular topic, with citations. Citizens who want to see how well their elected officials match their preferences. Anyone who wants to quickly and easily parse through previous board meeting statements. What are the pain points? The current Granicus video player has poor subtitle support, and only vaguely links agenda items to timecodes. It makes it difficult to follow how an issue has moved through the government.

What did we build?

We used the speech-to-text service DeepGram to precisely get transcripts of what people say, and which speaker has said it relative to other speakers. We take the detailed transcripts, split it into chunks that fit into the model context window, and annotate it with when the chunk was said and who said it metadata. We encode the semantic meaning of those chunks of text using Sentence Transformers’ all-MiniLM-L6-v2, and store the chunks of text, the semantic meaning, and the associated metadata in chromadb. We then use Qwen-3-32-B via Cerebus, to use RAG, Retrieval-Augmented Generation, to search the semantic meaning of anything any board, commission, or other public local official has said that we have indexed, and play the associated clip of what they actually said.