DeepL has a "Legacy Tech" problem. Why I built a wrapper that beats them at their own game.
Hello Indie Hackers! 👋
I'm the maker of O.Translator.
I have a controversial take: The era of proprietary translation engines (like DeepL's core models) is ending.
For the last decade, their moat was their data and their private neural networks. But with the rise of LLMs (GPT-5, Claude, Gemini), that advantage has vanished overnight. In fact, it’s becoming a liability—they are stuck trying to monetize their old models, while we (indie devs) can just plug into the smartest models on the planet immediately.
The playing field has been leveled.
📉 Where the Giants Fail
If "Text Translation" is a solved commodity, why is translating a document still so painful?
Because LLMs are great at language, but terrible at structure. If you paste a complex PDF or an Excel sheet into ChatGPT, you get a mess.
Layouts explode.
Images shift.
Data logic breaks.
🛠My 10-Year Bet on "Document Engineering"
With 10+ years of development experience, I realized the opportunity wasn't to build a better translator—it was to build a better Engineer.
I built O.Translator to be the bridge between Structure and Intelligence.
The Engine (AI): We don't train models. We let you choose the best ones (GPT-5, Claude, Gemini). This ensures the text quality is always superior to traditional machine translation, rivaling human accuracy.
The Wrapper (My Code): I spent months writing custom parsers that deconstruct documents into pure structure, feed only the text to the AI, and then reconstruct the document pixel-perfectly.
The Result?
Complex PDFs: We handle multi-column layouts and embedded elements that usually trip up standard converters.
Word Docs (DOCX): We don't just replace text; we traverse the underlying XML tree. This means headers, footers, complex tables, and custom styles survive the translation intact. No more spending hours fixing broken margins.
Excel Files: We don't just translate cells; we parse the AST to preserve formulas (VLOOKUP, SUM) and chart references. The logic stays, only the language changes.
💰 Better than Humans, Cheaper than SaaS
By combining Commoditized Intelligence (LLMs) with Specialized Engineering, we hit the sweet spot:
Quality: Higher than traditional MT (thanks to LLMs).
Speed: Instant.
Cost: A fraction of human agencies.
We operate on a "Pay-as-you-go" model with a Free Preview. You can see the part translated document layout for free. You verify the engineering quality before you pay a cent.
🧪 Discussion
I'm curious to hear from other devs:
Do you think proprietary models (like DeepL) can survive against general LLMs in the long run?
What’s the most annoying document format you’ve ever had to parse?
Roast the tool here: https://otranslator.com
Thanks!
