How Semantic Search Works

BrainTube uses a sophisticated multi-phase search system that goes far beyond simple keyword matching.

Hybrid Search

BrainTube combines two search approaches:

• Semantic search (70%) — Understands the meaning of your query using vector embeddings
• Keyword search / BM25 (30%) — Matches exact terms and phrases

The Search Pipeline

For complex queries, BrainTube uses a 6-phase Graph-Enhanced Search:

1. Query Routing — AI classifies your query to determine the best search strategy
2. BM25 (Keyword) — Fast text matching against titles, summaries, and transcripts
3. PageIndex — Checks pre-indexed important items
4. Semantic + HyPE — Vector similarity enhanced with Hypothetical Passage Embedding
5. GraphRAG — Traverses the Knowledge Graph to find connected content
6. CRAG (Corrective RAG) — Validates and re-ranks results for accuracy

Search Tips

• Natural language works best: "What did I learn about React hooks?"
• Quotes for exact matching: "useEffect cleanup"
• Entity names for people/tools: search by name
• Broad concepts for exploration

Search vs. Brain Chat