🔍 Browser-based Vector Search

×

About Browser-based Vector Search

A standalone browser demo of semantic search using vector embeddings. All processing happens in your browser with no backend or database required.

🔧 Technologies

• Transformers.js: Browser-native ML via @huggingface/transformers
• Embedding Model:
• In-Memory Search: JavaScript-based cosine similarity
• Static Hosting: GitHub Pages

📦 Embedding Preparation

• Structure-based chunking of Markdown files using AST parsing
• Semantic sections preserved with configurable chunk size & overlap
• Pre-computed embeddings generated via GitHub Actions workflow
• Stored on GitHub in the data-main branch

⚙️ How It Works

• Pre-computed embeddings loaded from GitHub at runtime
• Your query is embedded in-browser using Transformers.js
• JavaScript computes cosine similarity to find top matches
• No server, no database - everything runs locally

🔀 Retrieval Modes

• Hybrid (default): Fuses dense semantic search and BM25 lexical search via Reciprocal Rank Fusion (RRF) for best overall recall
• Dense (Semantic): Cosine similarity over pre-computed vector embeddings — best for conceptual queries
• BM25 (Lexical): Exact-match keyword search — best for filenames, symbols, error messages, and rare terms

📚 Learn More

• View webapp source
• About the embedding model

🏷️ Build Info

• Commit: