Translation quality & format preservation: - Word: merge adjacent same-format runs into one unit (sentence-level coherence like inline-tag handling); translate comments/balloons; dedupe textbox collection (was translated twice); RTL no longer overrides center/justify alignment; CJK/Arabic font hints (eastAsia/cs) - PPTX: chart translations now actually reach the output file (ChartPart.blob is read-only — rewrite chart XML in the saved ZIP); CJK typeface hints (a:ea) - Excel: sheet renames no longer break references — rewrite cell formulas (3D/quoted), defined names, data validations, cond. formats - PDF: bold/italic honored (hebo/heit/hebi); table cells never merge; unchanged blocks left untouched (typography preserved, fixes duplicate hyperlinks); attempted/changed stats + route gate now cover PDF; CJK font paths; scanned PDFs via Mistral OCR (detection + admin settings) Features: - formality param (formal/informal) + automatic regional-variant prompts - output_mode=bilingual docx (source above translation) - per-user translation memory on Redis (falls back to LRU), context-hashed - QA report + 0-100 confidence score in job status; L0 on by default - OpenAI-compatible providers: whole chunk in ONE numbered-JSON request (~15x fewer calls) with per-item fallback; base prompt always present (custom prompt no longer replaces translation instructions) Infra & marketing alignment: - plan-based engine gating + vision gating (closes paid-engine leak); /providers/available filtered per plan; 107 languages exposed - zh-CN/zh-TW validation fixed; libmagic disabled on Windows (native crash) - admin: Mistral OCR settings + engine status dashboard; httpx<0.28 pin (TestClient breakage); Prometheus test fixture fixed - marketing docs aligned with code (PDF+OCR, retention, engines, pricing) - security: .env.ionos/.env.production/provider_settings.json removed Tests: 1173 passed / 0 failed (6 network tests deselected: free Google endpoint temporarily blocked from this machine)
2.5 KiB
Hacker News — Show HN
Statut : prêt à publier (Phase 2, J0 matin EST). Soumettre via « Show HN » — pas de self-upvote.
Titre
Show HN: Office document translation that preserves the formatting
Corps
I built Office Translator (Wordly.art) because every document translator I used — Google, DeepL, the one-shot tools — destroyed the layout. Translating a 50-page Excel pricing matrix meant getting back a text dump: merged cells gone, formulas replaced by computed values, slide decks unrecognizable. The translation was 5% of the job; fixing the format was 95%.
So I made the format-preservation the product:
- In-place translation for .xlsx / .docx / .pptx / .pdf. The pipeline parses the native structure (cells + merges + formulas, paragraph runs + styles, slide XML + shapes), extracts only translatable content, translates it, and writes it back into the same structure. Output looks like it was authored in the target language.
- Scanned PDFs work too: image-only pages are recovered with OCR (Mistral) before translation — most competitors (DeepL, Azure) reject those outright.
- 7 engines, user's choice per document: Google (free), DeepL, Google Cloud, OpenRouter LLMs in two tiers (DeepSeek / Gemini / Claude), OpenAI, and Grok (xAI). Cheap engine for drafts, premium LLM for client deliverables.
- Custom glossaries to lock technical/legal/medical terminology across documents.
- Vision translation for text inside images (paid plans).
- Privacy: uploads deleted after 30 min, results auto-deleted within 2 hours, zero data retention, content never used for training.
- API on the Business plan (10k calls/mo) — the web workflow is fully mirrored: submit file, poll job, download.
Stack: FastAPI (Python 3.11), openpyxl / python-docx / python-pptx, PyMuPDF, Postgres + Redis, Stripe, Docker, Next.js 15 frontend.
Free plan: 2 documents/month, no card. I'd love feedback from people who've worked on document-format-preserving pipelines — what edge cases do I not handle yet? (PDF incl. scanned/OCR is already in — next candidates are IDML and CSV.)
Notes de publication
- Soumettre à 8–9 h EST, mardi–jeudi
- Répondre à chaque commentaire dans les 2 h ; rester technique, pas commercial
- Pas de « upvote my post » — c'est interdit sur HN
- Préparer 2–3 réponses techniques de poches : formule Excel + localisation (format de nombre), fusion de cellules + traduction, gestion des runs mixtes (gras/italique au milieu d'un paragraphe)