Skip to content

Video RAG for Lecture & Meeting Search

A video RAG stack that transcribes, diarizes, and semantically indexes long-form video for timestamped Q&A.

OpenAI APILlamaIndexFastAPIPinecone
Abstract indigo and amber cover illustration for the Video RAG project, showing timestamped video slices threaded by amber retrieval beams.

Highlights

  • Whisper-based transcription with speaker diarization and topic segmentation.
  • Timestamp-anchored chunks with jump-to-moment playback in the UI.
  • Query rewriting for jargon, acronyms, and speaker-scoped questions.

Outcomes

  • Cut review time on 1-hour meetings to under 4 minutes on average.
  • Powered searchable archives across 12k+ hours of internal video.

Stack

  • OpenAI API
  • LlamaIndex
  • FastAPI
  • Pinecone
  • PostgreSQL