Skip to content

002-embedding-model.md

Latest commit

 

History

History
15 lines (11 loc) · 674 Bytes

002-embedding-model.md

File metadata and controls

15 lines (11 loc) · 674 Bytes

ADR-002: Embedding Model Selection

Status

Accepted

Context

We need an embedding model for converting document chunks into vectors for retrieval. Options considered: OpenAI text-embedding-3-small, local sentence-transformers (all-MiniLM-L6-v2), Cohere embed-v3.

Decision

Use OpenAI text-embedding-3-small as the default, with a provider abstraction that allows swapping models. For local development and testing, use a mock provider that returns deterministic vectors.

Consequences

  • Requires an API key for production use
  • Mock provider enables cost-free development and CI
  • Provider abstraction adds a small layer of indirection but enables flexibility