Vector Search Comparison Dashboard
Compare the effectiveness of standard full-article embeddings vs paragraph chunking.
Semantic Search
Unlike traditional keyword search, semantic search uses AI to understand the meaning of text. It translates text into high-dimensional coordinates (vectors). When you search, the system finds results that are mathematically closest to your query in "concept space."
Ingestion & Search
Ingestion happens ahead of time: the AI (Gemini) reads your stories, generates vectors, and saves them to a database (Postgres).Search happens in real-time: your query is instantly turned into a vector to find the nearest matching database entries.
Full vs. Chunked
Full-Article embeddings capture the average meaning of a whole story, which dilutes specific details.
Chunking breaks stories into paragraphs and embeds them individually. This preserves fine details and allows the search to pinpoint highly specific concepts.
🥤 The Smoothie Metaphor (Explain Like I'm 5)
Imagine taking a giant 5-course meal (a whole story) and throwing it into a blender. If someone tastes the resulting smoothie and asks, "Are there carrots in this?", it's really hard to tell because the carrot flavor is diluted by the steak, potatoes, and cake. That's a Full-Article embedding.
Now imagine putting every single ingredient into its own separate little bowl. If someone asks for carrots, you can instantly point to the exact bowl holding just carrots. That's Chunking!