Understanding Your Knowledge Corpus
How saved content becomes searchable, interconnected knowledge in BrainTube.
Your knowledge corpus is the backbone of BrainTube — a living, growing network of everything you've learned.
What Is a Knowledge Corpus?
Think of it as a compiled knowledge base. Unlike a simple bookmark list, BrainTube extracts full text/transcripts, enriches with AI-generated metadata, embeds content as vectors, connects items via a Knowledge Graph, and synthesizes insights across sources.
The Enrichment Pipeline
When you save content, it flows through: Save → Extract → Enrich → Embed → Connect → Cluster
Stage 1: Extract
Raw content is pulled from the source — transcripts from videos, text from articles.
Stage 2: Enrich
AI (Gemini 2.5 Flash) extracts summaries, key entities, core thesis, hypothetical search questions, and topic tags.
Stage 3: Embed
Content is converted to 768-dimensional vectors using OpenAI's text-embedding-3-small with Contextual Retrieval — a metadata prefix is added before embedding for better search precision.
Stage 4: Connect
The auto-linker creates Knowledge Graph edges for items sharing 2+ entities, items with >85% embedding similarity, and resolved entity synonyms.
Stage 5: Cluster
Items are grouped into topic clusters using vector similarity. Clusters with 5+ items become Second Brains.
Corpus Health
Your corpus health depends on enrichment coverage, embedding coverage, graph density (goal: <25% orphan rate), and freshness (attention decay).