All work

AI and media automation

AI Video Subtitle Translation Pipeline

I built a self-hostable video-processing workflow that transcribes Turkish speech, translates subtitles into a selected language, and produces downloadable SRT files plus video with rendered captions.

Role
Independent Product Engineer
Period
2025

Open-source AI workflow

From submitted video to reviewable subtitle outputs

Simplified workflow

Video submission

Queued background job

Speech transcription

Subtitle translation

Workflow derived from the public implementation; providers and outputs remain explicit and reviewable. Diagram moving from video submission to a queued job, transcription, translation, subtitle rendering, and downloadable SRT and video outputs.

Context

Open-source personal project

Role

Independent Product Engineer

Product type

AI and media automation

Core capability

Python

Product context

Turning separate, failure-prone media steps into one job flow

Video transcription, translation, subtitle generation, and caption rendering each have different runtimes and failure modes. The product needed a reviewable flow with visible progress, provider choice, useful error states, and downloadable intermediate and final outputs.

Attribution anchor

What I was responsible for

I published the complete initial application: browser interface, FastAPI job API, queue and worker architecture, media pipeline, Docker setup, deployment guidance, and operational documentation.

Selected scope

Selected contributions

  1. 01

    Built the FastAPI job API and browser interface for URL submission, language and provider selection, progress, errors, and results.

  2. 02

    Used Redis and RQ to move long-running transcription, translation, and media work out of the web request lifecycle.

  3. 03

    Connected video download, Whisper transcription, multiple translation options, subtitle generation, and FFmpeg rendering into one pipeline.

  4. 04

    Added polling, shared result storage, downloadable video and SRT outputs, Docker Compose services, and deployment and safety documentation.

Evidence, not reconstruction

System view

A simplified view of verified responsibilities and system relationships. It does not reproduce a private or historical interface.

Open-source AI workflow

From submitted video to reviewable subtitle outputs

The public processing path represented as an operational job flow rather than a fictional interface.

Simplified workflow

Video submission

URL, target language, and explicit translation-provider selection.

Input

Queued background job

Redis and RQ isolate long-running work from the API request.

System

Speech transcription

Media acquisition and Whisper-based Turkish speech transcription.

Process

Subtitle translation

Selectable translation providers with progress and error handling.

Process

Subtitle generation and render

SRT generation and FFmpeg caption rendering.

Process

Reviewable downloads

Subtitle file and captioned video available as separate outputs.

Output

Verified relationships

  • Video submissionQueued background job
  • Queued background jobSpeech transcription
  • Speech transcriptionSubtitle translation
  • Subtitle translationSubtitle generation and render
  • Subtitle generation and renderReviewable downloads
Workflow derived from the public implementation; providers and outputs remain explicit and reviewable. Diagram moving from video submission to a queued job, transcription, translation, subtitle rendering, and downloadable SRT and video outputs.

Engineering judgment

Selected technical decisions

Decision 01

Separated the API and worker so expensive media jobs do not block request handling and can expose durable progress.

Decision 02

Kept translation-provider choice explicit instead of hiding cost and capability differences behind one fixed path.

Decision 03

Exposed subtitle files and intermediate progress so model output could be reviewed rather than treated as unquestionable.

Technology context

  • Python
  • FastAPI
  • Redis
  • RQ
  • OpenAI Whisper
  • PyTorch
  • yt-dlp
  • FFmpeg
  • OpenAI API
  • deep-translator
  • Docker Compose
  • HTML
  • CSS
  • JavaScript

Delivered capability

An end-to-end, self-hostable AI-assisted media workflow—from submitted URL through queued processing to reviewable subtitle and rendered-video files.

AI workflow automation · Python services · Media processing

Have an AI, document, or media process that should become a reliable product workflow?