Undertow Engine
Long-form in. Short-form out. Auto-published. The clip factory that runs while you sleep.
A standalone Python microservice that turns long-form content into a steady stream of published short-form video — without anyone sitting in front of a timeline editor.
What it is
Undertow Engine is the end-to-end pipeline for automated short-form video production and headless social publishing. Point it at a source — a podcast episode, a long YouTube upload, a recorded webinar, a Twitch VOD — and it sources, clips, scripts, composites, and posts the resulting short to TikTok, Reels, Shorts, or wherever you're trying to be seen.
Under the hood: FastAPI for the control plane, Celery for the work queue, MoviePy for the video compositing, Playwright for headless social-network posting, and Supabase for the state and asset storage that ties it all together.
It runs as a Docker container on the same kind of EC2 you'd run anything else on. It scales horizontally by adding worker nodes.
What it does well
Finds the clip-worthy moments. It scans the source media — transcripts, audio energy, semantic shifts — and extracts the timestamps that are actually worth turning into a short. You're not telling it where the good parts are. It's telling you.
Writes the script around the clip. The selected moment is reframed through an AI-scripting layer that produces the hook, the on-screen captions, the title, the hashtags — all in the voice you've configured for the channel.
Composites the video. MoviePy does the cuts, the captions, the b-roll overlays, the safe-area framing for vertical, and the audio-ducking around the voice. The output is a finished, platform-ready file — not raw cuts you still have to clean up.
Publishes without a human. Playwright drives a real browser session for each target platform, handling the login state and the upload flow. The platforms don't know it isn't you.
Logs everything. Every clip has a lineage — source URL, source timestamps, script version, render parameters, publication URL — so you can audit what shipped and roll back when something needs to be pulled.
Who it's for
Podcasters and creators who treat short-form as a distribution channel but don't want to staff an editor.
Agencies and content shops running short-form for ten or fifty clients at once, where doing it by hand isn't a P&L that works.
Brands and media properties sitting on years of long-form archive that should be working harder than it is.
Educators and course creators who want every lecture, episode, or recording to also exist as fifteen Reels by Monday.
What it looks like in practice
A podcaster uploads a 90-minute episode at 8am. Undertow Engine queues the source, extracts seven clip-worthy moments, drafts seven scripts in the host's voice, renders seven vertical videos with auto-captions, and publishes them across TikTok, Reels, and Shorts on a staggered schedule across the next 72 hours. The podcaster did nothing. The clip-by-clip analytics arrive in a Supabase dashboard the next morning.
A B2B agency runs undertow against twelve client podcast feeds nightly. Every morning each client gets a fresh batch of shorts auto-posted to their accounts. The agency's "video team" is one Celery cluster.
What it's not
It's not a video editor. You don't sit inside Undertow Engine. It runs without you.
It's not a general-purpose content-generation tool. The pipeline is opinionated about short-form video specifically — vertical, captioned, hook-first, under sixty seconds.
It's not an AI-replacement-for-creators. The source material has to exist. undertow just makes sure every minute of it gets the distribution it deserves.
Status
Shipping. Deployed via ECR + SSM. Talk to us for a walkthrough.