Self-Hosted Automated Video Pipeline Powering Two YouTube Channels
Challenge
Operate two daily-output YouTube channels covering different finance markets — Finance Forex Recap for the US market and Nifty Sensex Daily for the Indian market — without any manual video editing and without paying for cloud infrastructure. Every step had to run unattended on a local workstation, and the same pipeline had to be reusable across both channels with different branding and market-specific data sources.
Approach
Built a fully automated end-to-end pipeline executing entirely on local hardware — market data and news aggregation, AI script generation, TTS audio synthesis, video assembly, thumbnail generation, and YouTube upload — designed so a single codebase drives both channels via per-channel configuration.
Implementation
Kokoro-82M TTS model running on CPU-only — no GPU dependency. Per-market content sourcing with deduplication so the same story is not covered twice per channel. Python video assembly with FFmpeg. Channel-specific branding (intros, lower-thirds, thumbnails) generated via HTML/CSS rendering and parameterised per channel. Daily scheduling via local cron-equivalent runs both channels back-to-back, one for US market close, one for Indian market close.
Results
Both channels live in production — Finance Forex Recap (USA) and Nifty Sensex Daily (India) — publishing daily with zero manual intervention per video. Zero cloud infrastructure cost; the only ongoing cost is the electricity consumed by the local machine.