AI-Powered Content Automation Pipeline
Outcome
End-to-end automated pipeline producing and publishing video content across Instagram, YouTube, and X with zero manual intervention post-approval.
Build a fully automated content pipeline that takes a topic and produces publish-ready short-form video content — complete with AI voiceover, background music, motion graphics, and platform-optimized captions — for simultaneous multi-platform distribution.
Each platform has different format requirements (aspect ratios, caption limits, hashtag strategies, pixel formats). The system needed to handle text-to-speech synthesis, programmatic video rendering, music bed selection and volume calibration, and publishing APIs — all without manual editing software in the loop. Budget was near-zero: no paid video editors, no SaaS subscriptions beyond AI APIs.
Designed a pipeline architecture: Topic → Brief → Script → EDL (Edit Decision List) → Scene Audio Generation → Programmatic Video Render → Platform-Specific Post-Processing → Multi-Platform Publish. Used Remotion for deterministic, code-driven video rendering. Built a SQLite-backed content database as the single source of truth for all video metadata, render state, and publish status. ElevenLabs for voiceover, with LUFS-calibrated background music per track. Automated caption generation via LLM with platform-specific formatting.
Twitter rejects certain pixel formats (yuvj420p) — built an ffmpeg post-processing step for color range conversion. Instagram requires publicly accessible video URLs during upload — integrated ngrok tunneling into the publish workflow. YouTube OAuth token management proved unreliable via SDK — implemented raw curl-based upload as a stable workaround. Music volume calibration required per-track LUFS analysis to maintain consistent background levels across different compositions.
Expanding into long-form content with scene-based composition system. Building a review dashboard for batch content approval. The architecture is designed to be content-agnostic — the same pipeline can produce different content verticals by swapping briefs and visual themes.