Yusuf Emre Genc

- Fine-tuned Llama2-13b and LeoLM-13b models for e-learning script generation. - Used BERT to predict special token positions and align them between inputs and outputs for coherence. - Employed Hugging Face Transformers framework.

- Developed real-time emotion recognition using MobileNet and EfficientNet B2. - Deployed model into video conferencing app.

Worked with Pardus Linux, shell scripting, ISO image building.

Built two-way XML–Excel Interface Control Document converter in Python for embedded systems team.

Built tomato disease classifier with EfficientNet using PyTorch, OpenCV, Tensorboard.

2.0/5.0 Master's Thesis: Repository-Level Code Generation Research - Developed semantic-tree-based hierarchical retrieval and code generation system using LLM-based retrieval and prompt engineering techniques. - Achieved 0.46 pass@5 on RepoExec benchmark (vs 0.32 baseline). - Outperformed dense RAG baselines using 13× fewer LLM calls.

3.7/4.0