All projects · AI Agents & RAG
AI-Powered Semantic Search Engine with Supabase pgvector & Vector Embeddings
An AI-powered semantic search engine and software guides studio deconstructing how YouTube and modern search find exact matches from conceptual…
An AI-powered semantic search engine and software guides studio deconstructing how YouTube and modern search find exact matches from conceptual intent queries. Powered by 384-dimensional dense vector embeddings, Supabase pgvector extension with HNSW indexing, and cosine distance search (<=>), featuring side-by-side Lexical vs. Semantic comparison, 2D latent space visualizer, and Reciprocal Rank Fusion.
- Answers how YouTube and modern AI search engines pull exact content when users type vague or colloquial phrases
- Transforms 114 technical software guides and user queries into normalized 384-dimensional dense semantic vector embeddings
- Implements Supabase pgvector cosine distance search (<=>) with HNSW graph indexing and match_semantic_articles RPC stored procedure
- Proves intent matching: searching 'broken API' successfully maps to Webhook retries, GraphQL schema errors, and REST circuit breakers with zero literal keyword overlap
- Interactive 4-way search modality switcher: AI Semantic (pgvector), Lexical Keyword (#48), Hybrid Search (RRF), and Side-by-Side Comparison
- Interactive 2D vector space scatter plot projecting 384-dimensional latent clusters (APIs, CRM, Scheduling, Databases, Security) with dynamic query rays
- Real-time similarity progress meters displaying cosine match percentages and distance metrics (<=>)
- Automated 8-point in-browser diagnostic test suite verifying vector normalization, cosine distance math, intent retrieval, and UI accessibility
Built with: AI Semantic Search, Vector Embeddings, Supabase pgvector, PostgreSQL 17, HNSW Indexing, Cosine Similarity, Reciprocal Rank Fusion (RRF), Latent Vector Space