Booq — Persian-first AI video translation and dubbing

At a glance

• Localising media meant a chain of translators, voice actors, studios and editors, each a separate supplier and bill.
• One Persian-first web app for translation, dubbing, podcasts, music, images and film, paid from a single wallet.
• Live and in active development, with customer release notes generated from production deploys.

Highlights

• Five ways to localise one video

  Original or translated subtitles, a dub, or a dub with either subtitle track; one output per language, plus a single bundle.

• A distinct voice for every speaker

  Multi-speaker dubbing detects who is talking, gives each speaker a distinct voice and aligns timing to the original.

• 30 voices and a podcast timeline

  Narration, silence and music clips on one timeline; the music ducks only while someone is speaking.

• Music and songs from a brief

  Original instrumentals and full songs, with a strict mode that keeps the user's exact lyrics and optional lyric translation.

• 115 visual styles in 6 groups

  Camera, colour, lighting, material, art style and technique, composed without prompt jargon; an optional higher-resolution copy of every image.

• Story to film in 9 guided stages

  Each stage shows its plan first; the user can accept it, regenerate, edit or upload their own. Character sheets keep characters consistent between shots.

The challenge

Localising a video usually takes a chain of specialists: a translator, voice actors, a recording studio and an editor, and often a designer and a composer as well. Each link is a separate supplier with its own file format and its own invoice, and the schedule slips in the handoffs between them.

For the Persian-speaking creators, businesses and educators Booq is built for, the conditions raise the bar further. Payments run in Toman through a local gateway, large video files travel over uneven connections, and the interface has to feel native in a right-to-left language rather than translated into one.

Underneath, every capability depends on outside AI providers, and each of them can fail or change independently. The platform had to keep working when one of them did, and the person with something to say should never need to become a video editor to say it.

Architecture

Booq is a set of independent services behind one web app, with a separate gateway for each AI provider. That one decision is what lets a model be switched, a key rotated or a failing provider bypassed automatically, without touching the rest of the product.

• Every job is queued. Heavy processing waits in durable priority queues with per-service limits instead of running inside a web request.

• Work resumes instead of restarting. Translations save checkpoints and pick up where they stopped after a crash, and large uploads resume on a poor connection instead of starting over.

• Money follows the work. One prepaid wallet in Toman, VAT shown up front, a permanent ledger, discounts applied exactly as quoted, and automatic refunds when a job fails.

• Three apps, not one. Customers, administrators and support staff each use a separately deployed app, and a sign-in from one does not work in another.

• Switches, not releases. Product areas are turned on or off without a release, and models and provider keys are changed from the admin panel.

The platform runs on a container cluster defined as code, with heavy workloads placed next to their data. Every release passes build, tests, a staging deployment and a smoke test before a person approves it for production, and an emergency rollback to the previous version can be triggered manually.

What I built

One workspace covers the whole path from source material to a finished file.

Video translation and dubbing

• Five output modes: original-language subtitles, translated subtitles, dub only, dub with original subtitles, or dub with translated subtitles.

• Multi-speaker dubbing: speakers are detected, each gets a distinct voice, and timing is aligned to the original.

• Automatic or interactive: run hands-off, or pause to review and edit subtitles before anything is rendered.

• Broadcast-ready finish: styled burned-in subtitles for vertical and widescreen video, with the original audio or background music kept softly under the dub.

• Batch and link intake: upload an archive of videos or paste a social-media link, with automatic source-language detection and the price shown before anything is charged.

Voice, podcasts and music

• 30 AI voices with tone direction, applied as an instruction rather than read aloud.

• Podcast Studio: narration, silence and audio clips on one timeline; the voice can enter over the tail of a music clip, and the music ducks only while someone speaks.

• Music and songs: original instrumentals from a brief, and full songs with AI or user lyrics, including a strict mode that keeps the exact words and optional lyric translation.

• Transcription to subtitles or plain text, with AI punctuation and optional human review.

Images and film

• Visual Style Studio: 115 styles in 6 groups, so a look is composed without prompt jargon, plus an optional higher-resolution copy of every image.

• AI Video Studio: a topic or story becomes a rendered film through 9 guided stages: story, breakdown, visual style, characters, character sheets, soundtrack, shots, edit and render.

• Sound design: 744 free public-domain sound effects inside the studio, on a multi-track audio timeline.

Around the work

• My Library keeps every upload and every output in one place, available wherever a file is needed.

• Private by default: only the owner can download an output unless they create a share link.

• Three interface languages: Persian by default, plus English and Arabic, with full right-to-left support, installable as a web app.

• A bilingual page builder and blog, so marketing pages ship without developers.

Outcome

Booq is live and in active development. Production deploys generate customer-facing release notes with internal names removed, translated into Persian and Arabic and editable by admins.

A creator can now take one video into several languages in a single project, as subtitles, a dub or both, and produce podcasts, music, images and films in the same workspace, with every result landing in one library. Each job shows its price, or its starting estimate, before it runs, and a failed job returns its charge automatically.

The operators can switch models, rotate provider keys and turn product areas on or off without a release, and publish marketing pages without a developer. Backups run hourly, and a production backup was verified by reading it end to end.

Post-mortem

Running the platform in production led to several changes of direction. Each one traded a comfortable assumption for a safer default.

Speech. Speech was first pinned to a single provider on purpose, because a different provider means a different voice. When that provider had an outage, dubs were assembled with silence where lines were missing. Speech now falls back through a chain of providers, on the principle that a slightly different voice beats silence.

Moderation. Content screening used to run inside project creation, so users waited between a finished upload and their project page. It now runs in the background, and a finished result is held at “content review in progress” until the review completes, so output is not presented as ready while its review is still running.

Releases and backups. Production deploys moved from automatic to a manual approval that only unlocks after staging checks pass. Backups moved to hourly, and the newest one was read end to end, because a backup nobody has ever read is not a backup.

My role

• As CTO, defined Booq's architecture: independent services, a separate gateway for each AI provider, and three separately deployed apps for customers, admins and staff.
• Took the platform from its first commit to a production container cluster defined as code, with release gates that end in a manual approval.
• Owned the product and pricing decisions in the launch programme (pricing campaigns, licensing and storage policy) and ran it as a structured programme with verification and rollback steps.
• Set the requirements for the AI Video Studio as product owner, and seeded the Visual Style Studio from his own curated style collection.
• Ran an AI-augmented engineering workflow, briefing coding agents to research best practice, plan in phases and ship through staging to production.

Engineering

• Providers fail; the product keeps working

  Each AI provider sits behind its own gateway with automatic fallback, and models and keys change from the admin panel without a redeploy.

• Built for crashes and weak connections

  Translations resume from checkpoints after a crash, and multi-gigabyte uploads resume on a poor network instead of starting over.

• Money handled defensively

  Failed jobs refund automatically, refunds are protected against paying out twice, and every transaction lands in a permanent ledger.

• Screened before it is shown as ready

  Images, video, audio and text are screened in the background, with an appeal route; a finished result is held until its review completes.

• Layered access control

  Separate apps for customers, admins and staff; the consoles require a second factor and use role-based permissions.

• Monitored, load-tested, backed up

  Dashboards with severity-routed alerts, pre-launch load tests against pass/fail thresholds, and hourly backups, one of which was verified by reading it end to end.

Outcomes

• Live and in active development, with a Persian-first interface in Persian, English and Arabic.
• A creator can take one video into several languages in a single project, as subtitles, a dub or both.
• Podcasts, music, images and films are produced in the same workspace and land in one library.
• Operators switch models, rotate keys and turn product areas on or off without shipping a release.
• Production deploys generate customer release notes, translated into Persian and Arabic and editable by admins.