Mohamad Sawan

Parts register / Part P4

Convo

My own product · in progress · Designer and developer · 2026

Scale model of the convo, held up in the arm's gripper
Fig. P4 Scale model, as the arm presents it on the home page. Rendered from the live scene.

Notes

Convo converts almost any file into any sensible other format, in bulk, in English or Arabic, with the whole interface mirrored right to left. I'm building it now: the stack runs end to end on my machine, and the next step is putting it on a server.

What it does

  • 1,416 conversion paths across 121 formats. 304 of them run in the browser in WebAssembly, so the file never leaves the machine. The other 1,112 run on the server, which deletes the files an hour after the job.
  • 63 tools: a PDF toolkit (merge, split, compress, watermark, sign, redact, protect, OCR to a searchable PDF or Word), image, video and audio tools, Arabic text tools, archives, and AI: transcription, translation, summaries, table extraction and chat with a document.
  • Arabic is first class: Arabic text is shaped correctly in every PDF, image and video it writes, and OCR and speech recognition read Arabic.

How it's built

  • One registry of formats, converters, options and tools, in YAML. The browser engine, the server and the web UI are generated from it, and it decides whether a conversion runs in the browser or on the server.
  • In the browser: a Next.js app that installs as a PWA, with converters in Web Workers and WebAssembly (FFmpeg, libheif, pdf-lib and more).
  • On the server: a FastAPI engine with resumable uploads, live progress and signed downloads, and Dramatiq workers, one container per toolchain (LibreOffice, Pandoc, Ghostscript, OCRmyPDF with Tesseract, FFmpeg, libvips, faster-whisper). Workers stay off the internet unless their job needs it.
  • Hosting: Docker Compose behind Caddy, built for ARM as well as x86, so the whole thing runs for $0 a month on a free cloud VM.