1. Read
The job ad is split into sentences, lower-cased, stripped of accents and stop words. Headings matter: whatever follows “Nice to have” or “Bonus” weighs half as much as what is required.
Lab machine learning you can try live
Does my CV pass your ATS?
Recruiters screen CVs with algorithms. I wrote my own, the other way round: paste your job ad, it reads it, holds it up against my background and returns a score out of 100 — with my strengths and my weaknesses. It doesn't flatter: it computes.
The score is computed in your browser. The job ad only goes to an AI if you ask for its review.
No written job ad? Jot down the role, the AI formats it
Your notes are sent to a free AI model (via OpenRouter), instructed to rephrase them without adding anything. The score is then computed here, by the algorithm.
0/100
Pluses
Minuses
The AI's review — neutral, written from the analysis above
Sends the job ad and the analysis to a free AI model (via OpenRouter). It can be wrong: the score itself comes from the algorithm.
The line of his background closest to your job ad
Show the maths
Under the hood no LLM: classic NLP, readable end to end
2. Spot
A lexicon of some forty skills, each with its French and English synonyms and my real level, from “no” to “every day”. What I don't have becomes a minus, in black and white.
3. Compare
TF-IDF over the job ad and every line of my background, then cosine similarity: the machine finds the line of my CV closest to your need, even when the keywords differ.
4. Score
A logistic regression combines skill coverage, similarity, domain (sport or not) and the experience gap. The weights? Hand-calibrated on a handful of reference job ads. Yes: artisanal overfitting. A real model would need labelled job ads — recruiters, you know where to find me.
JavaScriptNLPTF-IDFLogistic regressionRegression tests