Goftito
Goftito (goftito.ir) is a bilingual SaaS platform for translating documents with AI. It handles Word, PDF (including scanned files), PowerPoint, images, subtitles and plain text, and runs every translation through quality control before delivering a file that keeps the original’s formatting. A voice studio, a credit-based billing system and an Android app sit on top of the same platform.
ASP.NET Core
Project technology
Blazor
Project technology
Razor Components
Project technology
A closer look at the project
Goftito is an AI-powered translation platform that takes a document in any common format and gives back the same file, in the same layout, in another language. When a translation switches direction, for example English to Persian, the whole page is mirrored: text, tables, images and charts all move to the other side.
From the first need to the final solution
The initial need
Companies, researchers and students needed to translate catalogs, technical manuals, research papers and presentations between English and Persian (and other languages) without spending hours rebuilding the layout by hand. A translation agency is slow and expensive, and free tools break the document.
The main challenge
Machine translation is easy. Getting a usable document back is the hard part. Online translators return plain text or broken layouts, and right-to-left languages like Persian and Arabic make it worse: tables, images and numbers have to be mirrored correctly without breaking the file. On top of that, AI output is not always reliable. It can drop numbers, repeat sentences or leave words untranslated, and every model call costs money.
The proposed solution
A five-stage pipeline: the document is broken into structured segments, translated with OpenAI in coherent units, checked by an automatic quality-control step, repaired where needed, and rebuilt into the original format. Each format has its own engine. For PDFs, the original page is edited in place rather than redrawn, so vector graphics, colors and images stay intact. A translation memory and cache stop users from paying twice for the same sentence. Around this sit a credit and package system with Zarinpal and Zibal payments, a full admin panel with a separate pricing test panel, a voice studio (text-to-speech, voice cloning, voice changer, speech-to-text, music), and a bilingual, SEO-ready public site.
The project development process
Discovery & Product Definition
I mapped who needs document translation and what breaks in existing tools, then defined the core promise: the output file must look like the original. I set the supported formats, quality tiers and a credit-based pricing model.
Architecture & Translation Pipeline
I built a layered architecture and a five-stage pipeline: parse, translate, quality check, repair and render. Segments that belong together go to the model as one unit, so the translation reads naturally while each piece keeps its place in the layout.
Format Engines & RTL Mirroring
Each format got its own engine: Word, PDF (including scanned pages with OCR), PowerPoint, images and subtitles. The hardest part was right-to-left mirroring: tables, images, charts and footnotes move to the opposite side, while logos and photos are never flipped.
Quality Control, Self-Healing & Cost
An automatic review catches missing numbers, repeated passages and untranslated words, and repairs only the faulty parts, so jobs finish on their own. Translation memory, caching and request batching cut the number of AI calls and keep the price fair.
Product Experience: Panels, Payments & Voice Studio
I designed a warm, bilingual user panel with light and dark themes, a complete admin panel, and a separate test panel for trying new prices safely. Users buy credit packages through Zarinpal or Zibal, and a voice studio adds text-to-speech, voice cloning, speech-to-text and music generation.
Testing, Security & Launch
Every change was replayed against a corpus of real documents and covered by more than 2,200 automated tests. After a full security review, the platform went live on a Linux server with encrypted automatic backups, a bilingual SEO-ready public site, and an Android app.
Selected images and videos
Project Image
Project Image
Project Image
What this project delivered
A live, production platform at goftito.ir that delivers translated Word, PDF, PowerPoint, image and subtitle files in their original layout, with full right-to-left mirroring. It is covered by more than 2,200 automated tests, runs on a Linux server with automatic encrypted backups, and is also available as an Android app.
Let’s build a clear and effective digital experience.
From technical content and visual communication to websites, interactive systems, and automation.