🗓️ 07082025 1037

VECTOR SEARCH

Vector search is a technique for finding similar items based on numerical vector representations (aka embeddings) of data, instead of doing keyword or exact-match lookups.

🔍 Why Use Vector Search?

Traditional search:

• Looks for exact keyword matches
• Doesn't understand meaning or context Vector search:
• Finds items that are semantically similar, even if the words used are different

💡 How It Works (Simplified)

1. Convert data to vectors (embeddings)
   • e.g. a sentence like "I love pizza" → [0.3, -0.7, 0.1, ...]
2. Store all vectors in a special database or index
3. When you query (e.g. "best food"),
   • it’s also converted into a vector

🧭 Common Use Cases

Use Case	Description
🔍 Semantic Search	Search documents/images by meaning, not exact words
🧑‍🤝‍🧑 Recommendation Systems	Find similar products/users based on embeddings
🧠 AI Chatbots / RAG	Retrieve relevant knowledge chunks before answering
🖼️ Image Search	Find visually similar images
🧬 Genomics	Compare DNA embeddings

🛠️ Tools / Frameworks for Vector Search

Tool	Description
FAISS (Meta)	Fast indexing of vectors (C++/Python)
Annoy (Spotify)	Approximate Nearest Neighbor in Rust
Milvus / Qdrant	Scalable vector DBs with APIs
Weaviate	Full-featured vector DB with modules
Pinecone	Managed vector DB service
Elasticsearch + kNN	Vector plugin for hybrid search

✅ Pros

• Understands semantics, not just keywords
• Enables fuzzy, context-aware search
• Great for unstructured data: text, images, audio, etc.

⚠️ Cons

• Slower than keyword search (but getting faster!)
• Needs preprocessing: embedding generation
• Scalability and freshness challenges with large data

🧠 Vector Search vs Keyword Search

Keyword Search	Vector Search
Based on	Exact words	Semantic meaning
Example Query	"red shoes"	"comfortable running gear"
Finds	Pages with "red shoes"	Pages about sneakers or running shoes
Under the hood	Inverted index	Vector similarity (ANN)

---

References

• ChatGPT