sds-converter-core

A Rust library for bidirectional conversion between Safety Data Sheet (SDS) documents (Word/PDF) and the Japanese Ministry of Health, Labour and Welfare (MHLW) standard JSON format.

Supports documents in Japanese, English, Simplified Chinese, and Traditional Chinese.

日本語 | 中文

Looking for the CLI? Install sds-converter instead.

Features

• SDS document → JSON: Extracts text from PDF/DOCX and converts it to the MHLW SDS data exchange format v1.0 via LLM API.
• JSON → DOCX: Generates a JIS Z 7253-compliant 16-section Word document from the standard JSON, with localized section headings.
• Multilingual: Handles source documents in ja / en / zh-CN / zh-TW.
• Extensible LLM backend: Ships with Anthropic Claude, OpenAI GPT, and Google Gemini backends. Bring your own by implementing LlmBackend.

Installation

[dependencies]
sds-converter-core = "0.1"

Library Usage

Convert SDS document to JSON (Anthropic Claude)

use sds_converter_core::{
    converter::{AnthropicBackend, LlmConfig},
    convert_to_json, ConvertConfig, Language,
};

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let backend = AnthropicBackend::new(
        std::env::var("ANTHROPIC_API_KEY")?,
        LlmConfig::default(),
    );

    let config = ConvertConfig {
        source_language: Some(Language::Japanese),
        output_language: Language::Japanese,
        ..Default::default()
    };

    let (sds, warnings) = convert_to_json(std::path::Path::new("input.pdf"), &backend, &config).await?;
    for w in &warnings { eprintln!("WARN: {w}"); }
    std::fs::write("output.json", serde_json::to_string_pretty(&sds)?)?;
    Ok(())
}

Convert JSON to Word document

use sds_converter_core::{convert_from_json, ConvertConfig, Language, SdsRoot};

fn main() -> anyhow::Result<()> {
    let json = std::fs::read_to_string("output.json")?;
    let sds: SdsRoot = serde_json::from_str(&json)?;

    let config = ConvertConfig {
        source_language: None,
        output_language: Language::Japanese,
        ..Default::default()
    };

    convert_from_json(&sds, std::path::Path::new("result.docx"), &config)?;
    Ok(())
}

OpenAI GPT or Google Gemini backend

use sds_converter_core::{OpenAiCompatBackend, LlmConfig};

// OpenAI GPT
let config = LlmConfig { model: "gpt-4o".into(), max_tokens: 8192 };
let backend = OpenAiCompatBackend::openai(std::env::var("OPENAI_API_KEY")?, config);

// Google Gemini
let config = LlmConfig { model: "gemini-2.0-flash".into(), max_tokens: 8192 };
let backend = OpenAiCompatBackend::gemini(std::env::var("GEMINI_API_KEY")?, config);

// Any OpenAI-compatible endpoint
let backend = OpenAiCompatBackend::new(
    "api-key",
    LlmConfig::default(),
    "https://your-endpoint/v1/chat/completions",
);

Extract raw text from a document

Use extract_text to pull the raw text out of a PDF, DOCX, or plain-text file without making an LLM call. Useful for building custom pipelines or inspecting what the LLM receives.

use sds_converter_core::extract_text;

#[tokio::main]
async fn main() -> anyhow::Result<()> {
    let text = extract_text(std::path::Path::new("input.pdf")).await?;
    println!("{text}");
    Ok(())
}

Supported extensions: .pdf, .docx, .xlsx, .txt.

Validate an extracted SdsRoot

validate checks the structural completeness of an SdsRoot and returns a list of warning strings. It does not hard-fail — partial results remain usable.

use sds_converter_core::{validate, SdsRoot};

fn main() -> anyhow::Result<()> {
    let json = std::fs::read_to_string("output.json")?;
    let sds: SdsRoot = serde_json::from_str(&json)?;
    let warnings = validate(&sds);
    if warnings.is_empty() {
        println!("OK");
    } else {
        for w in &warnings { eprintln!("WARN: {w}"); }
    }
    Ok(())
}

Custom LLM backend

Implement the LlmBackend trait to use any LLM provider:

use sds_converter_core::{LlmBackend, SdsError};

struct MyLlmBackend { /* ... */ }

impl LlmBackend for MyLlmBackend {
    async fn complete(&self, system: &str, user: &str) -> Result<String, SdsError> {
        // Call your LLM API and return the raw JSON string response
        todo!()
    }
}

JSON Format

The output JSON conforms to the MHLW SDS Data Exchange Format v1.0 (厚生労働省SDS情報交換のための標準的フォーマット, published 2025-03-31).

The schema covers all 16 sections of JIS Z 7253 with ~200 structured fields.

{
  "Datasheet": {
    "IssueDate": "2024-03-31",
    "SDS-SchemaVersionNo": "1.0"
  },
  "Identification": {
    "TradeProductIdentity": {
      "TradeNameJP": "サンプル製品"
    },
    "SupplierInformation": {
      "CompanyName": "株式会社サンプル",
      "Phone": "03-0000-0000"
    }
  }
}

Language Support

Requirements

• Rust 1.75+
• An LLM API key (for convert_to_json only)
  • Anthropic: Get API key
  • OpenAI: Get API key
  • Google Gemini: Get API key
• Input files must be text-based PDF or DOCX
  • Encrypted PDFs are not supported (text extraction will fail)
  • Scanned/image-only PDFs are not supported (no text to extract)

References

• MHLW — SDS Standard Data Exchange Format (official page) (Japanese)
• SDS Data Exchange Format Developer Manual (PDF) (Japanese)

License

Licensed under either of:

• Apache License, Version 2.0
• MIT License

at your option.