Recursive

Recursive is a cutting-edge local-first desktop application designed for translating large PDF documents with a focus on accuracy and semantic validation. Built using Java 21, this application leverages on-device AI to translate and verify documents, ensuring that the output preserves the original meaning, intent, and critical nuances that can often be lost in standard machine translations.

Key Features

• Thorough Verification Process: Recursive employs a unique recursive translation loop that translates text, checks it against the original for accuracy, and re-translates as necessary. This repeated cycle helps catch errors such as missed negations, incorrect terminology, and improperly conveyed meanings.

• Comprehensive Error Checks: The application systematically validates the translation based on crucial parameters, including meaning, intent, numbers, units, proper nouns, references, and relationships. This rigorous process prevents critical errors that could lead to misunderstandings in important documents:

  Check	What it Protects
  Meaning & Intent	Ensures instructions, warnings, and prohibitions survive translation
  Negation	Preserves must and must not across retries
  Numbers	Ensures exact matches for quantities, dates, and versions
  Units	Keeps measurements intact; never converts them
  Proper Nouns	Names, product identifiers, and URLs remain unchanged
  References	Maintains section numbers and figure references
  Relationships	Ensures logical consistency in cause and effect conditions
  Terminology	Guarantees consistent technical translations across documents

How it Works

The architecture follows a clean design model, structured into five Maven modules that facilitate modularity and maintainability:

recursive/
├── recursive-domain/          // Contains entities, value objects, and ports
├── recursive-application/     // Dedicated to use cases and application logic
├── recursive-infrastructure/  // Manages integrations with libraries like PDFBox and Tess4J
├── recursive-app/             // The JavaFX presentation layer
└── recursive-packaging/       // Configuration and packaging assets

The application runs entirely on the user's machine, ensuring that sensitive documents are never uploaded to the cloud and maintaining user privacy.

Technology Stack

Recursive is built on a robust technology stack that includes:

• Language: Java 21 (utilizing virtual threads)
• UI Framework: JavaFX 21 (OpenJFX)
• PDF Processing: Apache PDFBox 3.0
• OCR Support: Tess4J (a wrapper for Tesseract)
• On-device AI: Ollama
• Persistence: SQLite for storage
• Micro-pipeline Engine: AdaptiveFlow

User Experience

The application features a user-friendly JavaFX interface, which provides users with various screens for managing translation tasks:

• Dashboard: Overview of system status and job progress.
• Jobs: Create and manage translation jobs.
• Translate: Engage in the translation pipeline, review and modify translated content.
• Models: Access and manage AI models locally installed.

The development of Recursive is ongoing, with several phases of deliverables already shipped, including a robust domain model and a fully functioning translation engine.

Conclusion

Recursive stands out as a powerful tool for professionals who require precise translations of complex documents, such as contracts and technical manuals, ensuring that every essential detail is captured accurately while operating within a secure, local-first framework.