lecture-gate: AI talks into my own notes, while I sleep

Description and use

lecture‑gate automatically processes new YouTube videos, creates cards with key conclusions and verifies quotes, then appends only new points to the author’s local notes, keeping them private.

Problem

Manually summarizing and adding AI engineering talks to personal notes is time‑consuming and error‑prone.

Outcome

81 talks processed, 388 claims, 346 quotes verified (89%). Notes grew from 42 to 44 lines in one day without altering existing content. The system handled server failures, timeouts, and duplicate merges reliably.

Workflow

• Detect new YouTube upload via push notification
• Server retrieves transcript via SerpApi
• Card Writer (Gemma 4 31B) creates card with conclusion, summary, and claims
• Quote Checker verifies each claim’s quote is word‑for‑word in transcript
• Concept Matcher (Gemma 4 31B + bge‑m3) groups similar ideas
• Category Keeper files talk into section
• Rule Tester validates rules
• Note Merger (Gemma 4 12B + EmbeddingGemma) appends new points to local notes
• Highlighter marks key phrases
• Notes Assistant answers questions on laptop

Evidence

The cards grow into one map of concepts, so ten talks about one idea become one page.

Source

Only those are ever shown. This checks that the speaker said it, not that the claim is right.

Source

Personal notes never leave the laptop: notes/ is gitignored and only Ollama reads it.

Source