Job Radar.

Why I made it

Job hunting is repetitive work: opening several sites, reading dozens of similar descriptions and discarding most of them. I wanted to receive only what makes sense for me, without spending hours scrolling through lists.

The idea

A radar that runs on its own, gathers jobs from different sources into a single format, separates the ones worth it and delivers everything organized in Notion.

With one important detail: what gets rejected is saved too, with the reason. Automatic filters make mistakes, and I want to be able to check what they threw away.

Creative process

Each job source returns data in its own way. The first step was normalizing: title, company, location, work model and link in the same format, no matter where it came from.

Then came the cost side. Sending every job to the AI for evaluation would be expensive and unnecessary, because a lot can be discarded with simple rules first. So the AI only steps in when a job has already passed the basic filters and really needs a closer read.

Decisions I made

• Normalization firstData in the same format keeps the rest of the workflow simple, no matter how many sources I add.
• AI only when neededRule-based filters first, the model after. Fewer calls, lower cost, same quality.
• Auditable rejectionsEverything that was rejected is logged in Notion with the reason, so I can review the filter.