Exercises: Score Embedding Model Candidates
Use starter.py to rank embedding model candidates by quality, cost, latency, privacy, and migration risk.
1 Score retrieval quality
Weight Recall@k and citation support higher than marketing claims.
2 Penalize operational risk
Account for latency, privacy, rate limits, and re-indexing burden.
3 Recommend a strategy
Choose local, managed, domain-specific, or hybrid based on scenario needs.
4 Produce a release note
Explain why the candidate wins and what must be validated before release.