vibe.review.retrieval.example_retriever¶
Few-shot example retrieval for requirement classification.
Retrieves relevant examples from the examples database to guide LLM classification with known-good cases.
Phase 2.3: Example Retrieval - Find parts reviewed for same requirement - Sort by similarity to target part - Ensure diversity: mix of positive and negative examples
RetrievedExample ¶
A retrieved example with relevance scoring.
ExampleSet ¶
Set of examples for few-shot prompting.
ExampleRetriever ¶
Retrieves relevant few-shot examples for requirement classification.
Uses a two-stage retrieval process: 1. Database query to find examples for the same requirement 2. Reranking by relevance to the current document excerpt
Selection strategy (per Phase 2.3): - Find parts reviewed for same requirement - Sort by similarity to target part - Ensure diversity: mix of positive and negative examples
Usage
retriever = ExampleRetriever(session) examples = retriever.get_examples( requirement_id="DORA-28.1", document_excerpt="Our ICT risk framework...", top_k=3, )
__init__ ¶
__init__(session: Session, rerank_provider: RerankProvider | None = None) -> None
Initialize the retriever.
| Parameters: |
|
|---|
get_examples ¶
get_examples(requirement_id: str, document_excerpt: str, top_k: int = 3, include_similar_requirements: bool = True, min_quality_score: float = 0.5) -> ExampleSet
Retrieve relevant examples for few-shot classification.
| Parameters: |
|
|---|
| Returns: |
|
|---|
get_diverse_examples ¶
get_diverse_examples(requirement_id: str, document_excerpt: str, per_classification: int = 1, min_quality_score: float = 0.5) -> ExampleSet
Retrieve diverse examples covering different classification outcomes.
Ensures the few-shot prompt includes examples of each classification type (yes, no, partial, not_applicable) when available.
| Parameters: |
|
|---|
| Returns: |
|
|---|