Manual review of meeting video removed: cited answers in seconds.
Collects and transcribes local government meetings into a searchable knowledge graph, so citizens and officials get answers in seconds, not hours of video.
Sector
Government
Timeline
8 Weeks
Work area
Knowledge and decisionsProject
Municipal Intelligence Platform
The problem
Local government decisions are made in public meetings, but the record lives in hours of video. Anyone who needed to know what was said about a topic, from staff to journalists to residents, had to find the right meeting and scrub through recordings by hand.
A civic data platform that ingests regional meeting archives, generates accurate transcriptions, and structures people, topics, and decisions into a knowledge graph. A RAG assistant (one that answers from the archive itself, with sources) lets users ask questions in plain language and get verified citations, removing hours of manual video review.
What we built
- 01Archive ingestionDurable pipelines that collect meeting recordings and documents from regional archives and keep them up to date.
- 02Meeting transcriptionAccurate transcripts with speakers and timestamps, so every answer can point to the moment it was said.
- 03Knowledge graphPeople, committees, topics, and decisions linked together, so related discussions across meetings can be explored.
- 04Cited answersA question-and-answer assistant that only answers from the archive and shows its sources, with evaluation to check citation accuracy.
Technology
- Vertex AI
- GCS
- Firestore
- Temporal
- GCP
Methods
- Data Ingestion
- RAG Evaluation
- Durable Workflows
How it works
A user types a question such as how a fee change was discussed. The assistant searches the transcripts and the knowledge graph, writes a short answer, and lists the quotes it used, each with the meeting, the speaker, and a timestamp that opens the recording at the right moment.
Manual work removed
- Scrubbing through hours of meeting video to find one discussion
- Manually noting who said what, and when
- Rebuilding the history of a topic across several meetings
Facing something similar? Let’s map it together.
Related
Next case · Operations
Operational Intelligence Platform
