Solo developer shipping AI and full-stack MVPs in 3-5 days | founder of CPC AI, Seoul
I run a one-person software studio in Seoul (CPC AI). I build and ship production-ready web/app and AI MVPs in 3-5 days, and I write about the pipeline behind that: stack choices, automation, and where AI agents actually help. Studio: https://cpcai-studio.netlify.app/en/
Markdown supported · ⌘↵ to post
Every boundary being a file on disk is the design choice I'd copy first. Being able to rerun step 4 alone with the artifacts from steps 1 to 3 is what makes an unattended pipeline something you can trust at 3 a.m. Your rule about no network calls inside the template is one I'd underline too. A font fetched at render time fails quietly, and you only notice when a video ships in a fallback typeface. One addition I'd suggest: put a hash of the template itself in the manifest next to the render hash, so when the layout changes you can tell which videos came from which version. Do you also pin the voice model version, so a provider update can't change how old re-renders sound?
Youngjun Yoon
The template hash is the one I would take, and honestly, it is not in there today. The manifest records the render hash, which tells me the output changed but not why — a layout change and a font fallback look identical in that field. Putting the template's own hash beside it means the manifest can answer "which videos came from the version before the tweak" without diffing files. Adding it before the next layout change is the cheap moment to do it.
On the voice model: it is not pinned, and I would not pretend the cache substitutes for pinning it. TTS is cached per text hash, so anything already shipped keeps its audio and a provider update cannot reach back and change it. What the cache cannot protect is the next render — if the model shifts between Monday and Friday, two videos from the same template can sound different and nothing in the manifest says so. Pinning the model id and recording it in the manifest is the same fix as the template hash, and it is the one I would add first.
Your font point is the one that costs a whole night: a silent fallback is the worst failure mode in this pipeline. That is why the render step runs with the network blocked and dies on the first frame rather than producing a video in the wrong typeface.