Under the hood

From nine hours of video to one cited answer

A retrieval pipeline sits between your question and the series. Six steps, no magic — every answer is traceable back to the second it was spoken.

1. 01

   Transcription & indexing

   Every episode of the series is transcribed and split into small, overlapping segments — each one pinned to its start and end second in the video.

2. 02

   Vector embeddings

   Each segment is embedded into a dense vector that captures the meaning of what was said, not just the words.

3. 03

   Semantic retrieval

   Your question is embedded the same way, and a similarity search pulls the handful of segments most likely to contain the answer.

4. 04

   Grounded synthesis

   A language model writes the answer from those segments alone — and abstains when the series never covered the topic.

5. 05

   Second-level citations

   Every claim carries a citation that deep-links to the exact second on YouTube, so you can hear the source say it.

6. 06

   Quiz generation

   The same transcripts seed interview-prep quizzes, graded per topic with clips to rewatch for whatever you miss.