zantetsu-core

ML-based anime metadata extraction and normalization engine.

Crates

• zantetsu - unified API surface
• zantetsu-core - parsing engine
• zantetsu-vecdb - canonical title matching
• zantetsu-trainer - training workflows
• zantetsu-ffi - Node/Python/C bindings

Features

• Heuristic Parser: Fast regex-based extraction for production use (~92% accuracy)
• Neural CRF Parser: Candle-based Viterbi decoder for maximum accuracy
• Quality Scoring: Configurable profiles for release validation
• Zero-copy: Sub-millisecond parsing with minimal allocations

Usage

use zantetsu_core::{HeuristicParser, ParseResult};

let parser = HeuristicParser::new().unwrap();
let result: ParseResult = parser.parse("[SubsPlease] Spy x Family - 01 (1080p).mkv").unwrap();

assert_eq!(result.title, Some("Spy x Family".to_string()));
assert_eq!(result.group, Some("SubsPlease".to_string()));

Supported Formats

• Sub-group notation: [Group] Title - Ep.ext
• Scene naming: Title.S01E01.1080p.WEB-DL.AAC2.0.H.264.ext
• Batch paths: .../Group/Title/Group Title - Ep.ext
• Multi-episode: Ep01-Ep05, Ep01-05
• Subtitle variants, FLAC, Blu-Ray, WEB, DVD, TV sources

License

MIT