Since Part 1, this series has been documenting the build of a distributed media platform from scratch — from a simple frustration (movies and anime scattered across five devices) through metadata extraction, transcoding pipelines, hardening, and finally the full production architecture in Part 5.
That architecture is no longer just a diagram. It's live.
What's Working Right Now
The MVP runs entirely inside Telegram — no app to install, no account to create. Just a bot.
Search — ask for a title, and the bot tells you whether it's ready to stream, still processing, or not in the system yet.
Playback — completed titles stream straight through the gateway/worker split documented in Part 5: a Koyeb API gateway handling the lightweight requests, an Oracle Cloud VM running the FFmpeg transcoding worker for everything heavy, connected through an Upstash queue.
Request flow — if a title doesn't exist yet, you can request it and it gets picked up for processing.
All of it running on free-tier infrastructure, same as the architecture from Part 5.
What's Still In Progress
Upload is the one piece not fully hardened yet. Pre-signed URLs mean the gateway never touches the video bytes directly — but the failure paths, duplicate handling at scale, and edge cases from real (messy) user uploads are still being worked through. Streaming and search are solid; upload is the next thing getting stress-tested.
Why This Post Exists
Every part of this series so far has been me testing against my own usage. This is the first point where that's not enough — I need people who aren't me actually hitting the bot and telling me what breaks.
If you try it, here's what I actually want to know:
Does playback feel instant, or is there lag you notice?
Is anything about how the bot responds confusing?
Does anything break outright?
Try It
I'll read every reply. Next up on the build side is the monitoring and logging layer teased at the end of Part 5 — actually seeing what the system is doing once it's running, instead of guessing from the outside. Upload hardening picks back up after that, and this round of feedback will help shape what needs the most attention when it does.
This is part of the ongoing Building a Distributed Media Platform from Scratch series.
Top comments (0)