AIAI/ML · Automation · Internal platform
AI short-form video agent
We built an in-house AI video pipeline that scripts, generates, edits and publishes short videos across several channels and platforms every day with no manual steps. Scheduling, failure recovery and performance analytics all run automatically.
Problem
Producing and posting short videos daily across several channels and platforms by hand does not scale.
Solution
A scheduler plans each channel's daily slots, an LLM (Claude Sonnet) writes the script, LTX generates the footage, and later stages stitch, upscale and add voiceover and subtitles before publishing to YouTube, Facebook, Instagram and TikTok with cross-post retries. A separate GPU pipeline renders map fly-through videos on a remote machine with readiness checks. A watchdog backs off failing channels, dead-letters alerts when Slack is unreachable, and daily analytics digests rank content by performance.
Key features
- Per-channel daily scheduling with time windows and minimum gaps
- LLM scriptwriting
- AI video generation, clip stitching and upscaling
- Voiceover and subtitles
- Automatic publishing to four social platforms with retry queue
- GPU map fly-through rendering on a remote machine
- Watchdog with back-off and dead-letter alert queue
- Daily summaries and analytics digests in Slack
Tech stack
- Claude Sonnet
- Google Earth
- LTX
- Python
- SSH
- TTS
- TikTok
- Whisper
- YouTube
- cron
- nvidia-smi
Services behind this project
Read more
22SEPBusiness process automation: where to start
Business process automation works best when you start small. How to pick the first process, map it, choose tools and build automations that fail safely.
AI & Automation5 min read