language-detection-expert

Language detection identifies the language of input text using statistical analysis and entity recognition. This Skill combines langdetect with Polish entity hints (PESEL/NIP patterns) to classify text and return a language label, confidence score, and detection method for routing to appropriate processing pipelines.

Yes. Language detection informs which redaction model to use by identifying the text's language first. This Skill routes multilingual data to language-aware branches—heuristic, semantic, or LLM—based on detected language and confidence, enabling more accurate PII handling across Polish and other languages.

Statistical detection (langdetect) analyzes character patterns; hybrid detection adds entity-based hints like Polish PESEL and NIP recognition. This Skill combines both methods, improving accuracy on short texts and multilingual content by anchoring statistical results with Polish-specific entity patterns before routing.

The Skill exposes a POST /detect endpoint accepting text input and returning language, confidence, and detection method. These outputs route requests to one of three downstream branches (heuristic, semantic, LLM) in Vigil Guard, enabling language-aware processing at scale.

This Skill requires Python 3.11, Flask 3.0.0, and langdetect 1.0.9, plus custom Polish entity patterns for PESEL/NIP recognition. These dependencies enable hybrid detection and 3-branch routing for PII and multilingual workflows on Vigil Guard v2.0.0.

Statistical detection struggles with short texts, code-mixed content, and language-specific identifiers. This Skill adds Polish entity recognition to handle these edge cases, but confidence scores remain lower when neither statistical nor entity signals are strong—check returned confidence before routing critical redaction tasks.