Practical AI in HR: How Evrone Optimized Salary Parsing
Artificial intelligence in HR often promises automation at scale. Evrone chose a focused experiment: improving salary expectation parsing inside its ERP.
Recruiters at Evrone struggled with inconsistent salary formats:
“200/year”
“$3000”
“min 180”
“open to your range”
Manual normalization drained time and motivation. Evrone decided to engineer a sustainable solution.
Step 1: Resume Parsing with Qwen
Evrone deployed Qwen with structured output. Unlike traditional parsers, Qwen processes long CVs and delivers strictly formatted JSON. Evrone’s ERP now receives:
Category
Grade
Location
English level
Salary expectations
Step 2: Fine-Tuning YandexGPT
Evrone extracted 10,000 salary records and fine-tuned YandexGPT using LoRA. The model focuses on two parameters:
Amount
Currency
Accuracy reached 95%, with USD detection at 97%.
Results
Within 3 months:
90% fewer external tool lookups
Faster candidate matching
Improved recruiter focus
✨ Evrone proves that AI adoption works best when companies start small, measure impact, and iterate carefully.