Repository navigation
馃煛 [QUALITY] RAG context has no limits or diversity聽#17
Description
Problem
Current context building:
context = '\n'.join([
f'--- Documento: {m.document_title} ---\n{m.chunk.content}'
for m in matches
])Issues:
- No token/length limit (context explosion)
- No diversity (might get 5 chunks from same doc)
- No metadata for citation (chunk_index, source_ref)
Solution
Create ContextBuilder in application layer:
class ContextBuilder:
def __init__(self, max_tokens: int = 4000, max_per_doc: int = 2):
...
def build(self, hits: List[SearchHit]) -> tuple[str, List[Citation]]:
'''Build context with diversity and limits.'''Acceptance Criteria
- Token/character limit enforced
- Max chunks per document
- Citations returned for transparency
Activity
FullFran commentedon Jan 26, 2026
Resuelto. Hay un ContextBuilder en con l铆mites de tama帽o (max_chars) y diversidad (max_per_document), adem谩s de citas con chunk_index y fuente. lo usa para construir el contexto. Criterios cubiertos, cierro la issue.
FullFran commentedon Jan 26, 2026
Corrijo comentario anterior (el backtick se interpret贸 en shell).
Resuelto. Hay un ContextBuilder en
src/services/context_builder.pycon l铆mites de tama帽o (max_chars) y diversidad (max_per_document), adem谩s de citas con chunk_index y fuente.src/services/rag_service.pylo usa para construir el contexto. Criterios cubiertos.