Semantic Stock Search

Find large-cap companies by describing them in plain language. For example, "parapapapaaa i'm lovin it" shows McDonald's. It's like Google Search for stocks, but based on meaning instead of keywords. Click a button below to see how it works:

How it works

We convert your query text into a 768‑dimensional vector (an "embedding", i.e., a list of 768 numbers) and then rank stocks by how close their vectors are to your query (using cosine similarity).

What is a text embedding?

An embedding is a list of numbers that captures the meaning of text. Similar texts produce vectors that point in similar directions (e.g., "Missile" and "Lockheed Martin" are relatively close). This website in particular uses Google’s gemini-embedding-001 model to turn text into a 768‑dimensional vector.

Cosine similarity diagram showing angles between vectors — Cosine similarity compares the angle between vectors. Smaller angle → higher similarity. This chart is in 2 dimensions, but vectors can have many more dimensions. Calculations are the same.

How a search works

Generate embeddings: For each company in our dataset (1,500 stocks), I wrote a description and sent it to the gemini-embedding-001 model to generate a 768‑number vector.
Embed your query: we send your query text to the gemini-embedding-001 model to get a 768‑number vector.
Compute similarity: for each company in our dataset, we calculate its cosine similarity with your query vector. This similarity value is the "score" in the search results and it ranges from -1 to 1.