Zero-Hallucination AI Translation for Long-Form Documents
ZiyuanLingo combines 1M-token contextual reasoning with deterministic three-tier glossary locking, citation integrity verification, and layout-preserving scholarly PDF export.
Under stochastic gradient Langevin dynamics, posterior sampling converges within $O(\epsilon^{-2})$ iterations[^4], as established in doi:10.1145/3456789.
在随机梯度朗之万动力学框架下，后验采样在 $O(\epsilon^{-2})$ 次迭代内收敛[^4]，该结论已由文献 doi:10.1145/3456789 严格证明。
The Licensee shall indemnify the Licensor within 30 calendar days of written notice, and may not assign sublicense rights.
被许可方应当赔偿许可方（自收到书面通知起 30 个日历日内），且不得转让分许可权利。
Built for Documents Where One Wrong Word Matters
Generic chat translators lose context across chapters, drop footnotes, and drift on technical terms. ZiyuanLingo enforces state-backed document invariants from ingestion to typeset export.
Three-Tier Glossary & Entity Memory
Synchronizes Organization Global Glossaries, Project Entity Registries (acronyms, character patronymics, institutional names), and dynamically sliced batch context envelopes.
3-Level Deterministic + LLM QA Gate
Every batch passes hard invariant checks (paragraph ID bijection, footnote markers, uncorrupted DOIs), cross-lingual numeric/date alignment, and an independent Critic Agent semantic audit.
Layout-Preserving Scholarly PDF Engine
Parses two-column research papers, cross-page footnotes, tables, and vector formulas—exporting publication-ready monolingual or side-by-side bilingual PDFs without breaking typography.
Domain-Calibrated Registers
Dedicated profiles for STEM & Humanities Research, Cross-Border Legal Contracts (strict shall/may/must mapping), Technical Manuals, and Historical Monographs.
Crash-Resilient Batch Checkpointing
Translate 400-page books and regulatory filings reliably. Atomic state snapshots track every unit from PENDING to TRANSLATED and VALIDATED.
Zero-Training Data Privacy
Enterprise and academic manuscripts remain strictly confidential. Customer uploads are encrypted in transit and at rest and are never used to train foundation models.
Engineered on Google Cloud & Vertex AI
ZiyuanLingo leverages Google Cloud's elastic serverless infrastructure and Vertex AI Gemini long-context models to process complex multi-hundred-page documents with sub-paragraph precision.
Structural AST & OCR
containerized Cloud Run workers parse PDFs, DOCX, and EPUBs into atomic source units while isolating LaTeX math, citations, and footnotes.
3-Tier Memory Retrieval
Matches domain glossaries and entity registries, assembling a context window with chapter synopses, preceding translations, and lookahead units.
Vertex AI Translation & Critique
Primary Gemini translation pass followed by an independent Critic Agent pass checking causal inversions, double negations, and terminology drift.
Invariant Gate & Export
Verifies 1:1 unit bijection, numeric equivalence, and DOI integrity before compiling clean bilingual PDFs, DOCX, or Markdown archives.
Flexible Plans for Individual Researchers & Global Teams
Start free in our interactive Web Studio and scale seamlessly as your document volume grows.
Explorer
For students & evaluation
- Up to 30 document pages / month
- Interactive Web Studio access
- Project glossary (up to 100 terms)
- Level 1–2 automated invariant checks
- Markdown & bilingual HTML export
Researcher Pro
For scholars, translators & legal counsels
- Up to 600 pages / month (PDF, DOCX, EPUB)
- Layout-preserving bilingual & monolingual PDF
- Unlimited 3-Tier Glossary & Entity Memory
- Full Level 1–3 Semantic Critic QA Gate
- Academic, Legal & Monograph domain profiles
Team & Lab
For research labs, law firms & publishers
- 3,000 pooled pages / seat / month
- Shared Organization Global Glossary DB
- REST API & CI documentation webhooks
- Dedicated Vertex AI batch priority queue
- Zero-retention compliance & audit logs