I specialize in bridging the gap between AI research prototypes and deployed software. Whether it's training robust Computer Vision models that generalize across unseen hardware, designing Retrieval-Augmented Generation (RAG) microservices, or engineering low-level streaming pipelines, my focus is on building resilient, high-performance ML systems.
- 📍 Location: Sakarya, Türkiye
- 🎓 M.Sc. Student: Information Systems Engineering at Sakarya University
- 🔬 Core Focus: Medical AI, LLM Agents, Real-Time Video Processing & Streaming MLOps
Languages :: Python, C++, C#, SQL
Frameworks :: PyTorch, TensorFlow, ASP.NET Core MVC, Entity Framework Core, FastAPI, Flask, Scikit-Learn
AI & Vision :: OpenCV, RAG, ChromaDB, Sentence Transformers, Multi-Modal Agents
MLOps & Ops :: Docker, GitHub Actions (CI/CD), Dask, Pandas, NumPy, Linux, Git, Multi-threading
| Project & Status | Highlights & Performance Metrics | Core Tech |
|---|---|---|
wbc-analyzer🚀 Live Demo 📄 Preprint |
Out-of-Distribution Medical Pathology AI. Custom DenseNet121 + WBCAttentionBlock + MedSwish achieving 98.53% in-distribution accuracy. Inference-time domain adaptation boosted OOD accuracy from 56.96% to 89.05% (+32.09 pp) without retraining. Includes GPT-4o & Gemini Grad-CAM clinical agent. | PyTorch OpenCV Flask Docker |
Computer Vision & Machine Learning Research Project Private Research (M.Sc.) |
Supervised M.Sc. thesis research. Real-time video processing with a focus on low latency and temporal consistency. Details withheld until publication. | Computer Vision Video Processing |
Computer Vision for Blood Cell Microscopy🔒 Research in Progress |
Blood smear image analysis toolkit. Chain from raw microscope images to quantitative cell measurements: physics-based colour and stain handling, stain normalisation across slides, nucleus and cell segmentation, texture and shape features, validated against expert annotations. Under advisor's supervision. | Python PyTorch Scikit-Learn OpenCV |
rag-project-assistant⚡ Live Microservice |
Portfolio RAG Microservice. Sentence-transformers + ChromaDB (L2 < 1.40 threshold gate) + Groq Llama 3.3 70B. Deployed on HF Spaces via Docker & FastAPI with IP rate-limiting. | FastAPI ChromaDB Llama 3.3 70B |
kinematic-action-recognition📈 0.9995 Macro F1 |
10 GB Streaming ML Pipeline. Out-of-core Dask ingestion, real-time ADWIN drift detection (81 windows/sec, 59 MB peak RAM), and LightGBM ensemble on motion-capture sensor data. | Python Dask LightGBM ADWIN |
popcorn-wagon🎬 Hybrid Engine |
Movie Recommender System. Combines content-based filtering (TMDB API) and collaborative filtering (MovieLens SVD) with Spotify Annoy sub-millisecond similarity search. | Python Spotify Annoy SQLAlchemy |
📫 Let's Connect: emirhan0yildirim@gmail.com • emirhanyildirim.me • LinkedIn