Feel Free To Explore My Projects!
Ghost Coder
Deployed a Qwen 2.5 LLM locally as a coding assistant, integrated via a custom web extension (HTML/CSS/JS). Enabled low-latency code generation and debugging with offline privacy for developer workflows.
Used Technologies:
HTML, CSS, JavaScript, Unsloth, Python
Ghost Reader
Developed a full-stack sentiment analysis tool that dynamically plays music matching the emotional tone of novels, using NLP models fine-tuned on literary datasets. Achieved 96% accuracy within 1,000 training steps as part of Google's Kaggle Gen AI Intensive capstone.
Used Technologies:
Transformers, Longformer, Unsloth, Python
Ocean Heatmap Visualization
Developed in a Kodluyoruz Ideathon project using synthetic RGB heatmaps (R: wind, G: waves, B: depth) to train a CNN for offshore wind site classification. Achieved scalable, threshold-adjustable analysis without manual data processing.
Used Technologies:
pytorch, CNN, Python
Ghost Agent (AI Customer Service Agent)
Developing a RAG-powered assistant (Gemini API + Node.js/JS) that analyzes user sentiment and wait time to automatically escalate frustrated customers to human agents. Combines real-time emotional tone detection with queue optimization.
Used Technologies:
Gemini API, Node.js, JavaScript, HTML, CSS
Colon Cancer AI (CNN Diagnostics)
High-accuracy convolutional neural network (TensorFlow/Keras) that detects malignant colon cells in medical imagery, achieving 98% training accuracy – enabling faster, AI-assisted pathology screening.
Used Technologies:
TensorFlow + Keras (CNN architecture), Python
Connect (Autism Support AI)
CNN-based emotion detection system (TensorFlow/Keras) to help individuals with Autism Spectrum Disorder interpret facial expressions in real-time. Trained on CK+48 and FER2013 datasets, the model classifies 7 emotions (anger, happiness, etc.) with 90% accuracy, and was prototyped for integration with smart glasses.
Used Technologies:
Gemini API, Node.js, JavaScript, HTML, CSS