How RAG Finds the Right Document Fast: Graph-Based Vector Search

Bashiri Smith · Facebook reel · 2026-09-25 · 1:08 · 12,921 views · Open on Facebook

Topics: Embeddings & Vector Databases, Retrieval-Augmented Generation (RAG), Resume, Job Search & Interviews · Level: intermediate

Summary

A skit about an interview question: how does a RAG app find the right document for a question? Embedding every document and comparing the query to every vector is a full linear scan, which is too slow at 10 million documents. The better answer is a layered nearest-neighbor graph (the HNSW approach, though the video doesn't name it). The search starts at a sparse top layer, takes big hops toward the query, then drops down to denser layers and moves in smaller steps. It only checks a few hundred vectors, so search time grows roughly logarithmically instead of linearly.

Key points

• Step 1 of a RAG retriever: embed every document (or chunk) as a vector.
• The naive answer, comparing the question's vector to every document vector and taking the best match, is a brute-force full scan. That's O(n) and too slow for 10 million documents.
• The answer interviewers want: an approximate nearest-neighbor index built as a graph, where each vector links to its nearest neighbors.
• Add a few sparse 'express' layers on top of the base graph so the search can make long jumps.
• Search starts at the top layer with big hops toward the query, then moves down a layer at a time, taking smaller hops each time.
• At each step, the search jumps to whichever neighbor is closest to the query (greedy search). Most vectors are never visited.
• One search touches only a few hundred vectors instead of millions, so linear search becomes roughly logarithmic.
• This layered-graph design is HNSW (Hierarchical Navigable Small World), used by most vector databases. The video describes it but never says the name; the name is added here so you can look it up.

Resources mentioned

• BASWE.Ai Engineer (Skool community) · community · skool.com · paid
  The creator's paid community and program, with an AI learning roadmap (including the full ops and evaluation track), daily calls with engineers and recruiters, resume and portfolio help, and a job-search pipeline.
  Also in: Basic RAG Pipeline in 60 Seconds: From Documents to Grounded Answers (Bashiri Smith on Facebook · notes), Pointer to Bashiri Smith's Complete AI Engineer Roadmap for 2026 (Bashiri Smith on Facebook · notes), Step-by-Step Roadmap to a $200K+ AI Engineering Role (Bashiri Smith on Facebook · notes), How to Evaluate a RAG Pipeline: Retrieval vs. Generation (Interview Answer) (Bashiri Smith on Facebook · notes) and 76 more

Try this

• Prepare to explain why brute-force vector comparison doesn't scale and how approximate nearest-neighbor graph indexes avoid a full scan.
• Practice explaining the layered graph search: big hops at the top layer, smaller hops lower down, greedy moves to the nearest neighbor.
• Optional (creator's pitch): join the creator's community for a learning plan, mock interviews and coaching.