Why our CV score doesn’t use an LLM
The same CV got three different scores in three runs. That was the moment we stopped asking a model for a number.
Zaid Hijazi
1 min read
On this page
Key takeaways
- Use the model to read, and rules to decide.
- A score people can’t explain is a score they won’t trust.
- Measure consistency: same input, same output.
The problem with asking for a number
When a language model gives the score directly, small changes in wording change the result. Candidates can’t see why they lost points.
Read with a model, decide with rules
The model extracts structured fields. Weighted, documented rules turn those fields into a score. Every lost point has a reason.
Not every scoring problem needs an LLM.
The result
Scores became fully consistent, faster and far cheaper — and a flawless CV can finally reach 100.
- AI
- Evaluation
- Explainability
Was this article useful? Thanks for the feedback!
Zaid Hijazi
Software & AI Engineer in Riyadh. I write about building systems where mistakes are expensive.