ADR 0003: Pinecone 1536-Index

Context

We need to select a vector embedding model and index configuration.

Decision

1. Standardize on OpenAI text-embedding-3-small.
2. Configure Pinecone with 1536 dimensions.
3. Use Cosine Similarity as the metric.

Consequences

• Positive: High performance; low cost (0.00002 / 1k tokens); simple integration.
• Negative: Locked into the 1536-dimension architecture; migrating models would require a full re-embedding of the curriculum.

Back to Index