Vector Search (Unified DB) is a multi-namespace search layer built by James Murray to unify stories, songs, and poems into a single semantic index. It powers the AI search experience on AddictionTube, allowing users to explore creative recovery content with natural-language queries.

Instead of searching “exact titles” or browsing long lists, users can ask for themes like “relapse and hope in the winter” or “poems about early sobriety and fear”, and the system surfaces the most relevant content from all categories.

Unified Content Model

• Stories: Narrative pieces about addiction, recovery, families, and turning points.
• Songs: Lyrics and audio-linked content designed for recovery-themed music projects.
• Poems: Spoken-word and written pieces that capture fine-grained emotional states.

Each item is turned into an embedding and enriched with metadata (type, tags, themes, mood, length) so the system can retrieve by both meaning and filters.

Search Experience

• Semantic Queries: Users ask questions or describe feelings; the system finds the closest matches.
• Mixed Results: Results may include a blend of stories, songs, and poems in one view.
• AI Answer Layer: A RAG endpoint can generate a short summary or reflection using the top results as context.

Technical Architecture

• Embeddings: Text-embedding model for all content, with optional per-type fine-tuning.
• Namespaces: Separate vector namespaces for stories, songs, poems, unified via a routing layer.
• Backend: Python RAG service + PHP proxy; JSON responses rendered by a custom search UI.

Try the Unified Search

To see the unified search interface in action, visit:
AddictionTube Vector Search

For organizations interested in building similar multi-namespace vector search for their content libraries, James offers consulting and implementation support. See Speaking & Workshops or Contact for next steps.