CV | Wonbeom Jang

2024 -
LLM Research Engineer

SK Telecom

A.X Foundation Model Development & LLM Benchmarking
- Designed and constructed post-training datasets for large-scale model training
- Developed proprietary in-house benchmarks to evaluate model performance
- Developed Korean language understanding train data and improved model performance
- Development of Telco domain-specific benchmarks
- Development of Korean safety benchmarks and LLM-as-a-Judge models
- Development and operation of in-house benchmarking systems
2023 - 2023
Deep Learning Research Intern

Neurocle
- OCR: Improved recognition performance while reducing training time
- Lightweight Anomaly Detection: Designed and implemented efficient anomaly detection models
- Smart Labeling Models: Converted and optimized models for smart labeling tasks
- Transfer Learning: Verified and evaluated transfer learning performance
2019 - 2020
Undergraduate Research Intern

CVML Lab. Chung-Ang University
- DPSGAN experiment and MDADN lightweight project were conducted

A.X Foundation Model
- Designed and constructed post-training datasets
- Designed and constructed Korean language understanding datasets
TelcoLLM, TelBench, TelAgentBench
- Achieved high agent satisfaction score
- TelBench: Designed and built Automatic Evaluation & LLM-as-a-Judge for Telco capability assessment
- Designed and built LLM-as-a-Judge evaluation with strong correlation metrics
- AICC deployment: LLM commercialized for AICC in-call RAG and post-call analysis
Smaller, Safer, Stronger: SK Telecom Tunes Gemma 4B for Multilingual Customer Support Moderation
- Achieved equal or superior performance in both Korean and English compared to proprietary models such as GPT-4.1 and Claude Sonnet 3.7
- Applied SFT + PPO reinforcement learning on small models including Gemma 4B and Llama 3.1 8B
- Outperformed proprietary large models with small open models