Practice Bank: Local Retrieval Decisions
Use starter.py to decide when TF-IDF, SVD, neural embeddings, or hybrid search should be used.
Scenario 1: Error code lookup
Exact lexical retrieval is strong.
Scenario 2: Credential paraphrase
TF-IDF may miss vocabulary mismatch.
Scenario 3: Offline demo
Local methods avoid network and API keys.
Scenario 4: Sensitive HR corpus
Local processing reduces provider exposure.
Scenario 5: Product SKUs
Token preservation is critical.
Scenario 6: Legal clause numbers
Sparse signals protect exact references.
Scenario 7: Many common words
IDF helps reduce noise.
Scenario 8: Low SVD dimension
Compression may blur important distinctions.
Scenario 9: New corpus release
IDF and projection may need rebuild.
Scenario 10: Hybrid production search
Local lexical and neural retrieval can complement each other.