Built a search tool to help me find a job but it failed silently

I decided to build a tool to help me find jobs of my choice.

My goal was simple. I wanted to find small companies with 1 to 50 employees, where reaching out directly to a founder might actually work. But searching for them on LinkedIn was frustrating. Selecting the various filters again and again.

I wanted to have a simpler and easy to use search. So, I decided to build one.

I wanted to keep the project as simple as possible. I used no complex code, no heavy tools, and no databases. I just wanted a simple web page where I could select a location, an industry, and a company size.

The tool was supposed to give me a web link with the parameters adjusted based on my query.

I used an AI tool to help me write the code faster. It created the initial design, wrote the logic, and set up checks to make sure everything worked. Very quickly, the tool looked ready.

Then I started using it, and reality set in.

The first problem was that my basic assumptions about LinkedIn were wrong.

I thought that if a filter for company size worked in one type of search, it would work everywhere. It did not.

When I tried to use my generated link to find small companies, LinkedIn did not show an error. It did not tell me anything was wrong. Instead, it quietly ignored my filter and showed me huge companies like Apple and Microsoft.

I realized that LinkedIn treats different searches as completely different systems. A filter for finding a person uses a totally different hidden label than a filter for finding a company. Even simple categories like Software had different hidden numbers behind the scenes depending on what I was searching for.

I stopped guessing. I logged into LinkedIn, clicked the filters by hand, and looked closely at the links created in my browser bar. Seeing what the platform actually did became my only source of truth.

That was when I noticed an even bigger problem. LinkedIn search links are fragile and unpredictable.

Some filters only work if you pay for a premium account. If you do not have one, LinkedIn quietly strips those choices out of your link without telling you. The link was unreliable and could break at any time.

By the time I stopped working on it, my code was clean and all my checks were passing. My code was doing exactly what I told it to do. But no automated check could tell me if LinkedIn was going to silently ignore my choices in the real world.

AI made writing the code very fast, but it could not replace real judgment. The AI was happy to write clean code around an assumption that was wrong. The real work was testing how the real world actually behaved.

Trying to build a reliable tool on top of a system you cannot control is very difficult. But the project was not a waste. It taught me a lot about testing real behavior, questioning simple assumptions, and knowing when a project has done its job.

Not every project needs to launch or turn into a business to be worth your time.