EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 21·5 min read
Show HN: ParqDB – Vector search in the browser from Parquet over HTTP"
TL;DR
ParqDB enables vector search directly in the browser using immutable Parquet files stored in object storage, eliminating the need for a query server. This innovative approach allows users to search through 100,000 Wikipedia articles efficiently and privately.
✦ Why It Matters
Explore ParqDB to enhance your web applications with efficient, private vector search capabilities.
Key Takeaways
- Implement ParqDB for private vector searches in web applications.
- Utilize MiniLM embeddings for efficient indexing of text data.
- Store search data as immutable Parquet files for better performance.
- Leverage WASM for fast ranking of search results in the browser.
How It Works
ParqDB builds an IVF-LVQ8 index over MiniLM embeddings, which are compact representations of text data. When a user types a query, the browser retrieves only the necessary byte ranges from the Parquet files stored in object storage.
The ranking of results is executed in the browser using WebAssembly (WASM), which allows for high-performance computation without sending data to a server.