TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Generative Mechanics · AGE 18—20 · FACTUAL · DURABLE

Vector similarity search

A folder finds what you named, and a vector search finds what you meant.

The idea

If every document, drawing and photograph has coordinates in a space where near means similar, then "find me projects like this brief" becomes a measurement. Which stored points sit nearest to the point of my query? That is vector similarity search. A vector database stores the coordinates of each item alongside the item, and answers nearest-neighbour questions quickly, even over millions of items. It is the basic operation that retrieval runs on. It lets a studio search its own archive by intent, climate or building type instead of by filename.

Why it matters

A folder finds what you named. A vector search finds what you meant, including things you named badly in 2019. That changes how you can use your own past work.

See it in the studio

"Narrow plot, hot-dry, courtyard, low budget" returns a 2017 house whose folder was called `Final_FINAL_v3`. The filename never said courtyard. The embedding did.

Watch for this

Thinking nearest means best. Nearest means nearest in the model's sense of similar, which may weigh a photograph's palette more than your brief's climate logic. Read the results. Do not trust the ranking.

Try it

Write three short descriptions of past studio projects. Ask a model which two are most similar and why. Then ask which two a client would call similar. The answers will differ, and the difference is what "similar" means in the space.

Prove it

Explain how "find projects like this brief" becomes a distance measurement in the space, and describe what a vector database stores that plain folders cannot.

How it works

Each item is embedded once and stored. A query is embedded the same way, and the database returns the items at the smallest distance (or the highest cosine similarity). Exact search over millions of items is slow, so real systems use approximate indexes that trade a little accuracy for speed. RAG on this strand is what happens next: the nearest pieces are handed to the model as context.

What this idea builds on

What this idea opens up

Sources

Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON