I am a Computer Engineering PhD candidate at Yıldız Technical University (expected 2025), combining cutting-edge NLP research with extensive production engineering experience in mobile and backend systems.

My work uniquely bridges AI research and real-world software engineering—from developing Turkish LLMs and medical AI systems to architecting scalable Flutter applications and cloud-based backends. I specialize in LLM fine-tuning, adaptive learning, and model merging for low-resource languages, while maintaining deep expertise in cross-platform mobile development and full-stack architecture.

Research Focus:

• NLP for Low-Resource Languages (Turkish, Kurdish, Arabic)
• Large Language Model Development (LLM training, fine-tuning, evaluation)
• Medical AI & Healthcare NLP (Patient-doctor Q&A, clinical text processing)
• Multilingual Tokenization (Morphologically-aware, semantic-preserving tokenizers)
• Benchmark Development (Turkish MMLU - 6,200 questions across 62 sections)

Engineering Expertise:

• Flutter & Dart (Cross-platform mobile development, state management, advanced architectures)
• Mobile AI Integration (On-device ML, TensorFlow Lite, intelligent mobile applications)
• Backend Systems (Firebase, Django, Node.js, scalable cloud architectures)
• Full-Stack Development (React, Angular, Swift for iOS-native features)
• DevOps & CI/CD (Docker, automated deployments, production pipelines)

International Experience:

• 🇵🇹 Visiting PhD Researcher, Instituto Politécnico de Tomar (Erasmus, 2023)
• 🇬🇷 MSc Exchange, Aristotle University of Thessaloniki (Erasmus, 2020)

Languages: 🗣️ Turkish, Kurdish (Native) | English, Arabic (Fluent) | Persian, Zazaki (Conversational)

📫 Contact: malibayram20@gmail.com | 🔗 LinkedIn | 🤗 Hugging Face

"Training on Real Patient-Doctor Q&A Data"

• Fine-tuned Llama 3 (8B) on 167k+ authentic patient-doctor dialogues
• Applied LoRA adapters & SLerp merges for knowledge retention
• Evaluated against GPT-3.5 with expert clinical feedback
• Developed Doctor-Llama & DoctorGemma model series

"Setting Standards in Turkish NLP: TurkishMMLU for Large Language Model Evaluation"

• Created comprehensive 6,200-question benchmark from 280k+ questions
• Covers 62 sections, 67 disciplines, 800+ topics
• Established first standardized evaluation framework for Turkish LLMs
• Improved tokenization quality and cross-domain semantic understanding

"Data Quality-Based Adaptive Learning Rate: A Case Study on Medical Text Classification" (Conference Paper)

• Developed dynamic learning rate mechanism tied to content quality
• Applied to 167k medical dataset, improving convergence and accuracy
• Enhanced efficiency for specialized NLP applications

• Tokenizer Projects: Linguistically-informed tokenizers for Turkish & morphologically-rich languages
  • github.com/malibayram/tokenizer
  • github.com/malibayram/tokenizer_benchmark
• Model Architectures: Attention mechanisms, embeddings, and performance profiling
  • github.com/malibayram/model-architectures

• Udemy & YouTube Instructor: Creating comprehensive courses on Flutter, mobile development, and AI
• Research Assistant: Yeditepe University - Teaching AI, programming, and ML fundamentals
• Mentorship: Leading Flutter, Android, and NLP bootcamps and developer communities
• Technical Content: Producing educational materials for thousands of learners worldwide

Mobile Expertise: Advanced Flutter architectures • State management (Bloc, Provider, Riverpod) • Native platform integration • Performance optimization • Firebase integration

AI Research: LLM Training & Fine-tuning • Transformers • LoRA Adapters • Model Merging • Tokenization • MMLU Benchmarking • Multi-GPU Training

Backend & Cloud: RESTful APIs • Microservices • Real-time databases • Cloud Functions • Scalable architectures • CI/CD pipelines

This section automatically updates with my latest content via GitHub Actions.

Automatically updated daily with WakaTime statistics.

"Bridging linguistic diversity and computational intelligence to build AI systems that understand and serve multilingual communities."