🗂️ AI & Agents · View mindmap

Vector Search

Vector search (or similarity search) finds semantically similar items by comparing high-dimensional vector embeddings. It is foundational for semantic search, recommendation systems, and retrieval-augmented-generation-rag.

Core Mechanisms

• Vector Embeddings: Numerical representations of data (text, images) in a continuous vector space (e.g., via BERT, Sentence Transformers).
• Similarity Metrics: Cosine similarity or Euclidean distance to measure vector proximity.
• Approximate Nearest Neighbor (ANN) Algorithms: Efficiently search large vector databases (e.g., FAISS, HNSW).

Limitations and Alternatives

• Embedding Consistency Requirement: Traditional vector search requires the same model for both embedding generation and retrieval (e.g., using text-embedding-ada-002 for both steps).
• Graph RAG: A flexible alternative using knowledge graphs and LLMs for structured retrieval. Does not require consistent models for embedding vs. retrieval, enabling more adaptable systems.
  • IBM Explainer: Creating GraphRAG (video demonstration)
  • Key advantage: Graph structures allow explicit relationship traversal (e.g., “find all products related to battery via manufacturer”).

Applications

• Semantic search in enterprise knowledge bases
• Personalized recommendations
• Contextual LLM augmentation via rag

2026 04 14 IBM Explainer creating GraphRAG

Source Notes

• 2026-04-14: “But OpenClaw is expensive…”
• 2026-04-07: Chroma Context 1 Self Editing Search Agent for Efficient RAG · ▶ source
• 2026-04-08: Structured AI Context Beyond RAG Limitations with Map First Architectu · ▶ source