Show Dev: I scored 16K LinkedIn profiles. Most of them land around 59/100

About a year ago I started a side project: paste a LinkedIn URL, get a 0-100 score for the profile and a list of what to fix. It's called LinkedOptimizer. 250 commits later it has scored 16,775 profiles, and the numbers are more interesting than I expected, so here they are, plus how the thing is built and a few things I broke along the way.

What the scores look like

The median profile scores 59. The average is 56.

• 23% score under 40
• 32% score 70 or higher
• you need 73 to be in the top quarter, and 82 to be in the top 10%

Each profile gets four section scores. The averages:

Experience is the weakest section, and it's the one the overall score weights most, because it's what recruiters spend the most time reading. That combination explains a lot of the low totals.

The fixes the scorer suggests most often are boring. Bullets that list duties instead of results. A headline that's just a job title. An About section of two sentences. Skills that don't match the role you're actually going after (you can give it a target role, and the score changes a lot when you do).

How it works

The stack is plain:

• Flask app on Google Cloud Run, Postgres, Redis
• It reads the public profile from the URL you paste, so nobody has to log in with LinkedIn
• Gemini scores each section against the target role and writes the rewrites
• Paddle for payments

The part that took the most debugging was rate limits. On busy days Gemini returns 429s. The first version just failed. Now it walks a list of models (several Flash versions) before it rotates API keys, so a 429 on one model usually means the request quietly lands on the next one.

Things I got wrong

I once shipped a typo in a model name. Every single analysis failed until I noticed. There was no alert for "100% of analyses failing", which says more about my monitoring than about Gemini.

I ran a price test on the paid report: $9 vs $4.95. $4.95 won, which surprised me. Round two ($4.95 vs $6.99) is running now.

I also put a heavy blur over the locked part of the report to push upgrades. Two days later I added the opposite: one real, unblurred example rewrite per section, because a blurred box gives people no reason to believe the paid part is any good.

Try it

The score and the diagnosis are free, no signup and no LinkedIn login: linkedoptimizer.ai. If you run your profile through it, I'd like to hear where the score felt wrong. That's the feedback I learn the most from.