Finding product ideas in freelance job posts.
MarketTrends read freelance job posts and found the ones that pointed to a product someone could build. We built the scraper that collected the jobs, the AI that sorted them, and the web app where people found ideas.
- Product
- Web app with paid plans
- Data
- Upwork job posts
- AI
- Claude Haiku and Sonnet
The problem
People keep paying for the same fixes.
Businesses post jobs on Upwork all day, and a lot of them ask for the same tools and fixes. When enough people pay a freelancer for the same thing, that’s a product someone could build.
Nobody has time to read every post. We built MarketTrends to read them for you and show the patterns.
What we built
A scraper, an AI sorter and the app on top.
The scraper (GLaDOS)
A Go service that pulled new Upwork jobs every six minutes. It kept its own login alive, cleaned up budgets, job types and skills, skipped jobs it already had, and saved everything to Postgres.
AI sorting
Claude Haiku read each new job and decided if it pointed to a product idea. If it did, it tagged the idea type, the problem, the industry and the kind of buyer. Claude Sonnet then double-checked the ideas and wrote daily reports for each category and industry.
Pattern detection
Ideas that kept coming up got grouped together and scored on how well they’d work as a software product.
The web app
A Next.js app where people browsed ideas, filtered by type, industry and problem, saved the good ones, and tracked trends and skills over time. Daily and weekly email digests kept them up to date. Pro users could export to CSV, and there was a REST API with personal keys.
Accounts and billing
Sign in with email, Google or X. Free, Starter and Pro plans with a 7-day trial, paid through Stripe. We also built an affiliate program and an iOS and Android wrapper with in-app purchases.
Where it ended up
Built in under two weeks, then paused.
The first commit landed on January 6, 2026. By January 15 the scraper, the AI sorting, the dashboard, Stripe billing and the API were all built.
The scraper and the app shared one database. GLaDOS wrote the jobs, and the app read them, sorted them and built the reports. No API between them, so there was less to break.
The project was paused in January 2026. The scheduled jobs were switched off, so the ideas stopped updating. The site is still up.
Want something like this built?
Tell us what you want to build. We’ll tell you how we’d build it.