Case study / Product engineering
VideoRead
VideoRead watches long-form video so you do not have to. It takes content from curated channels and turns it into clean, editorial-style articles using AI transcription and summarisation. Conceived, designed and engineered by ouibild as a working production system.
How it works
Video in. Article out.
Ingest
Videos arrive from curated channels and are queued automatically as they publish. No manual chasing.
Transcribe
Full transcripts are produced through a dedicated transcription pipeline with a fallback provider behind it.
Compose
A two-stage AI pass writes the article body, then a separate pass writes the headline, summary and metadata.
Publish
Finished pieces land in a clean reading experience, formatted like a quality publication, not a transcript dump.
01
The problem
There is more good long-form video published every day than anyone could ever watch. Most of it is locked inside a format you cannot skim, search, quote or read on a quiet train. The value is in there. The format keeps it out of reach.
VideoRead was built to unlock it: take the hours of video from channels worth following and turn them into articles you can actually consume, at the speed you read, without sitting through a fifty minute upload to find the five minutes that matter.
02
The engineering
Two-call AI architecture. Article body and metadata are generated in separate passes. Asking one prompt to write a thousand-word piece and produce clean structured metadata at the same time is how you get mediocre versions of both. Splitting the work keeps each output focused and reliable.
Duration-based model routing. A three-minute clip and a ninety-minute feature are not the same job. The system routes each video to the model best suited to its length and complexity, balancing quality against cost on every single piece.
Resilient sourcing. Transcription runs on a primary provider with a fallback ready to take over. A single supplier having a bad day never stops the pipeline or leaves the queue stuck.
Cost control as a first-class concern. Real AI products fail on the bill as often as on the output. Routing, batching and sensible limits keep the economics sane as volume grows.
03
The result
A running system that quietly turns a firehose of video into a readable library of articles, with no one sitting in the middle copying and pasting. It is the kind of project that shows what ouibild means by app development: not a brochure site with a chatbot bolted on the side, but a genuine production system with queues, fallbacks, routing and a handle on its own costs.
If you have a product idea that depends on AI being done properly rather than demoed convincingly, this is the proof that ouibild can build it and keep it running.
For ambitious teams
Got a product idea that needs AI done properly?
We have shipped this once, for ourselves. We can ship it for you.