Weaviate — Module 14.2

How This Lesson Fits the Module & Volume

Weaviate is an open-source vector database that combines vectors with a class/property object model, optional vectorizer modules, and strong hybrid (BM25 + vector) search—aligning with Module 14.1’s hybrid search ideas.

After Milvus’s scale focus, Weaviate emphasizes schema, modules, and GraphQL/REST developer experience for AI apps.

Learning Objectives

By the end of this lesson, students should be able to:

Describe Weaviate collections (classes), properties, and vector indexes.
Ingest objects with vectors or built-in vectorizer modules.
Run near-vector and hybrid queries with property filters.
Compare self-hosted Weaviate vs Weaviate Cloud.
Explain when modular vectorizers help vs hurt control.
Position Weaviate among FAISS, Chroma, Pinecone, Milvus, and Qdrant.

Definition

Weaviate is an open-source vector database that stores data objects with properties and vectors, supports semantic and hybrid search, and can attach modules (e.g., embedding providers) so objects are vectorized on insert or query.

Object Model + Vectors

Concept	Role in RAG
Collection / class	Type of document chunk or entity
Properties	Text + metadata filters (source, tags)
Vector	Bring-your-own or module-generated
Hybrid search	BM25 + vector fusion in one query
Modules	Optional embedding / generative helpers

Client Sketch: Insert and Hybrid Query

import weaviate
from weaviate.classes.config import Configure, Property, DataType
from weaviate.classes.query import Filter, HybridFusion

client = weaviate.connect_to_local()

# Example: bring-your-own vectors (no auto-vectorizer)
if not client.collections.exists("Chunk"):
    client.collections.create(
        name="Chunk",
        properties=[
            Property(name="text", data_type=DataType.TEXT),
            Property(name="source", data_type=DataType.TEXT),
            Property(name="tenant", data_type=DataType.TEXT),
        ],
        vectorizer_config=Configure.Vectorizer.none(),
    )

chunks = client.collections.get("Chunk")
chunks.data.insert(
    properties={"text": "PTO accrues monthly.", "source": "hr.md", "tenant": "acme"},
    vector=embedding_list,
)

result = chunks.query.hybrid(
    query="vacation policy",
    vector=query_embedding,
    alpha=0.5,
    fusion_type=HybridFusion.RELATIVE_SCORE,
    filters=Filter.by_property("tenant").equal("acme"),
    limit=5,
)
for obj in result.objects:
    print(obj.properties["text"], obj.metadata.score)

client.close()

Modules: Convenience vs Control

Module vectorizers

Faster prototypes
Provider keys in Weaviate
Less app-side code

BYO vectors

Full model control
Easier offline/air-gap
Consistent with FAISS pipelines

Hybrid

Keyword + semantic
Tune alpha / fusion
Matches 14.1 lessons

Strengths

Hybrid search first-class
Rich object/property model
Modules accelerate demos
OSS + cloud options

Tradeoffs

Schema design overhead
Module coupling to vendors
Ops for self-host clusters
API surface to learn

Common Misconception

“Hybrid alpha = 0.5 is always optimal.” Keyword-heavy SKU queries and paraphrase questions need different blends. Tune with labeled queries like any other IR hyperparameter.

Knowledge Check

Short Answer: What two signals does Weaviate hybrid search combine? Answer: BM25/lexical and vector/semantic.
True/False: Weaviate can store properties (metadata) with objects. Answer: True.
Multiple Choice: BYO vectors means: (a) Weaviate trains GPT, (b) you supply embeddings, (c) no vectors allowed. Answer: (b).
Short Answer: Why use filters on tenant properties? Answer: Multi-tenant isolation / ACL-style scoping.
True/False: Vectorizer modules are mandatory. Answer: False.
Multiple Choice: Alpha in hybrid search typically balances: (a) CPU vs GPU, (b) keyword vs vector weight, (c) batch size. Answer: (b).
Short Answer: Name one benefit of Weaviate vs raw FAISS. Answer: Object schema, hybrid search, filters, modules (any).
Short Answer: Name a deployment option. Answer: Self-hosted or Weaviate Cloud.
Multiple Choice: Module vectorizers trade convenience for: (a) zero cost always, (b) less control / provider coupling, (c) larger kernels. Answer: (b).
True/False: Hybrid search aligns with Module 14.1 hybrid retrieval ideas. Answer: True.

Key Takeaways

Weaviate models RAG data as objects with properties and vectors.
Hybrid BM25 + vector search is a core differentiator.
Choose modules for speed or BYO embeddings for control.
Tune fusion weights with real query slices.
Next: Qdrant—Rust-powered vector DB with strong filtering and payloads.

Trainer’s Guide

Lab: Insert the same corpus; compare pure vector vs hybrid at alpha 0.25 / 0.5 / 0.75 on keyword vs paraphrase queries.

Debate: Module vectorizer vs app-side Sentence-Transformers for regulated data.

Recap: Weaviate pairs object search with hybrid retrieval. Continue with Qdrant.