SECAI Core · phase 3 of 15
Vector storage
A lookup table that turns pieces of text into numeric fingerprints
The Explain card
- Plain English
- Vector storage is a lookup system that turns pieces of text into numeric fingerprints (embeddings) and finds the entries whose fingerprints are closest to a query.
- Example
- A retrieval-augmented chatbot stores every company policy as a vector. When a user asks about expense limits, the nearest vectors are fetched and handed to the LLM as context.
- Why it matters
- Whatever sits in the vector store becomes what the model "knows". Poisoned entries feed injected instructions into answers, and embeddings can sometimes be reversed to recover the original text.
- Hook
- A library sorted by meaning rather than by title.
Where it sits in the deck
Phase 3: Data Fundamentals: Types, Pipelines, and Preparation
Models are only as good as their data — understand the raw material and the engineering that shapes it before covering how it flows or breaks.
