> For the complete documentation index, see [llms.txt](https://oodlz-ai-studio.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://oodlz-ai-studio.gitbook.io/docs/inbound-agent/the-brain/query-preprocessing.md).

# Query Preprocessing

Filler phrases stripped deterministically, FAQ queries detected and passed through — vector search runs on what the user meant, not what they literally typed.

> **Strip the filler before the vector search runs — retrieve relevance, not noise.**

Generic AI assistants pass the user's raw message directly to a vector search. "Show me cashback deals on shoes" and "shoes" return very different results. Most tools never clean the query — so they retrieve noise instead of the most relevant brands. The Oodlz query preprocessing layer fixes this before any retrieval happens.

## How It Works

Every incoming query passes through a deterministic preprocessing layer before the vector search is called. Filler phrases are stripped — `deals on`, `cashback for`, `show me`, `looking for`, `can you find`, and a dozen more like them — leaving a clean noun-phrase that produces precise vector results. FAQ queries are detected separately by question-word analysis and passed through untouched, since stripping fillers from "how does cashback work" would corrupt the intent.

The cleaned query is what the vector search sees. The original user message is preserved in the session log for memory and review, but the retrieval call itself runs on the noun-phrase the user actually meant.

## What You Get

* Vector search runs on intent, not literal text — relevance instead of noise.
* FAQ queries handled separately so question phrasing is preserved.
* Same preprocessing applied to every channel — website, social, voice — so retrieval quality is consistent.
* Deterministic stripping — same input always produces the same cleaned query, so debugging a bad result reproduces reliably.