qfa.services.language#

Source-language detection for prompt assembly.

Wraps langdetect directly so the services layer can pin an LLM’s output language to the language a feedback record was written in. Deliberately separate from qfa.adapters.presidio_anonymizer.detect_language(), which answers a different question (which of Presidio’s six spaCy models to load) and would be an adapters import from the services layer besides.

Functions

detect_source_language(text)

Detect the language text is written in.

qfa.services.language.detect_source_language(text: str) → str | None[source]#

Detect the language text is written in.

Returns a raw ISO 639-1 code ("en", "fr"), or None when the text is too short, too noisy, or too ambiguous to detect reliably — callers are expected to skip pinning a language in that case rather than trust a guess. Never raises. Only the first 2000 characters are inspected.