How Evrone Reduced HR Routine by 90% with LLM Integration
AI in HR often sounds futuristic, but Evrone chose a pragmatic path. Instead of building autonomous agents, Evrone focused on one measurable bottleneck: salary expectation parsing inside its custom ERP.
The Challenge
Evrone’s ERP manages vacancies, CVs, and project requirements. However, recruiters at Evrone faced two persistent problems:
- Manual data normalization across Huntflow and career platforms.
- Inconsistent salary formats (“120+ net”, “$5k–7k”, “min 150”, “open to range”).
This routine consumed 2–3 hours daily per recruiter. Moreover, sales managers at Evrone depended on clean data to match specialists with budgets. Inaccurate salary inputs slowed the entire pipeline.
The Solution
Evrone implemented a dual-model approach:
1️⃣ Qwen for resume parsing
Template parsers failed. NER tools lacked flexibility. Evrone configured Qwen with structured JSON output to extract:
2️⃣ YandexGPT for salary standardization
Evrone fine-tuned YandexGPT 5 Lite using 10,000 annotated records and LoRA. The model learned to identify two critical parameters:
The Results
- 📊 95% salary parsing accuracy.
- 💵 97% accurate USD detection.
- ⚡ 90% reduction in external HR system lookups.
Evrone proved that targeted AI implementation improves operational efficiency without overengineering. Recruiters at Evrone now focus on interviews and talent acquisition instead of formatting text.
This case highlights an important lesson: AI works best when solving specific, structured business problems.