Struct ResponseClassifier 

pub struct ResponseClassifier;

Stateless classifier for LLM responses.

Implementations

impl ResponseClassifier

pub fn new() -> Self

Create a new classifier.

pub fn classify(&self, response: &str, prompt: &str) -> ClassificationResult

Classify a single response given the original prompt.

pub fn detect_category(text: &str) -> (ResponseCategory, f64)

Detect the response category using keyword/pattern matching.

Returns (category, confidence).

pub fn detect_refusal(text: &str) -> bool

Detect if the response is a refusal.

pub fn detect_citations(text: &str) -> bool

Detect whether the response contains citations or references.

pub fn score_coherence(text: &str) -> f64

Score sentence-level coherence (0-1).

Heuristic: rewards smooth sentence beginnings, penalises abrupt stops.

pub fn score_completeness(response: &str, prompt: &str) -> f64

Score how completely the response addresses the prompt (0-1).

pub fn score_conciseness(text: &str) -> f64

Score conciseness (0-1). Higher = more concise.

pub fn score_formatting(text: &str) -> f64

Score formatting quality (0-1).

pub fn flesch_kincaid_grade(text: &str) -> f64

Compute the Flesch-Kincaid Grade Level.

FK Grade = 0.39 * (words/sentences) + 11.8 * (syllables/words) - 15.59

pub fn batch_classify( &self, responses: &[(String, String)], ) -> Vec<ClassificationResult>

Classify a batch of (response, prompt) pairs.

Trait Implementations

impl Debug for ResponseClassifier

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ResponseClassifier

fn default() -> ResponseClassifier

Returns the “default value” for a type.