The problem
A deals brand lives or dies on volume and trust. It needs fresh deals every day, posted in several places, and every deal has to be real. DailyPrimeStore's promise is that it checks the price history so its followers don't fall for fake discounts.
Done by hand, that is hours of searching, checking, designing and posting, every day. We wanted to see how much of it software could do well, and where a person still needed to look.
What we built
A pipeline of scheduled jobs and AI agents, where each stage hands its output to the next:
- Research. Agents look for products worth posting.
- Price checks. Each candidate is checked against its price history, and only real drops move forward.
- Image and video rendering. Post graphics and short videos are generated for each deal.
- Multi-platform posting. Deals publish to the website and to social platforms on a schedule.
- Weekly email. The best deals of the week are rolled into an email.
The website is built with Astro and served from Cloudflare Pages. GitHub Actions runs the scheduled jobs, so every run is logged, repeatable and easy to rerun when something upstream fails.
What it shows about building automation
Running our own pipeline every day shapes how we build automation for clients:
- Verification beats generation. The valuable step is the check that throws out bad deals, not the step that writes the caption.
- Every run should leave a trail. When a source changes or a post fails, the logs show exactly where and why.
- People own the rules. What counts as a good deal is decided by a person; the repetitive work goes to software.
- Cheap, boring infrastructure wins. Static hosting and scheduled jobs keep the running cost and the failure surface small.