| Parameter | Value |
|---|---|
| Current Epoch | MSc in ECE · Abdullah Gül University · Est. 2027 |
| Loss | Converging ↘ (slowly, like all good research) |
| Status | 🟢 Active · Multiple experiments running in parallel |
I'm Erol Külüşlü, a Software Engineer and MSc Researcher in Computer Vision & Applied ML at Abdullah Gül University, currently working as an AI Specialist at Kayseri Bilişim Akademisi.
I build and study end-to-end intelligent systems — spanning self-supervised multimodal learning, graph neural networks for biomedical AI, retrieval-augmented generation, and production-grade software engineering.
Highlights:
- Designing CM-JEPA: a self-supervised RGB-IR fusion framework using JEPA-based latent prediction, VICReg regularization, and gradient-based structural consistency.
- Adapting MuloAD (GraphSAGE + VCDN) for colorectal cancer classification from gut microbiome data — best result: cross-population AUC = 0.808 (JPN → IND).
- Built MNEME: a temporal knowledge graph RAG system over 187 documents across four domains — submitted for journal publication.
- Designing and delivering AI/ML training programs for university students, teachers, and public institutions.
- Built production NER pipelines, OCR benchmarking frameworks, and serverless AWS infrastructure across 2.5 years at ILA Amsterdam.
- Writing on Medium about LLMs, context engineering, and applied AI systems.
- International experience: Erasmus+ exchange in the Netherlands + volunteering across 10+ countries.
Building systems that understand context, retain knowledge, and make intelligence accessible — at every scale.
|
Computer Vision RGB-IR multimodal fusion |
Biomedical AI Graph neural networks for drug repositioning |
Knowledge & Retrieval Temporal knowledge graph RAG |
|
Optimization Dynamics Adaptive learning rate methods |
AI Systems & Agents RAG pipelines · LLM orchestration |
Software Engineer End-to-end Flutter apps · BLoC · OAuth2 |
experiments = {
"CM-JEPA": {
"desc": "Cross-modal RGB-IR fusion via JEPA-based latent prediction",
"loss": "L_JEPA + α·L_VIC + γ·L_grad",
"eval": ["YOLOv8-Nano", "MobileNetV2/U-Net"],
"data": ["M3FD", "MSRS", "LLVIP", "TNO"],
"status": "🟡 Active Research",
},
"MuloAD-CRC": {
"desc": "GraphSAGE + VCDN for colorectal cancer · gut microbiome",
"result": "Cross-population AUC = 0.808 (JPN → IND)",
"key_biomarker": "Peptostreptococcus stomatis",
"status": "🟠 Preparing Journal Submission",
},
"MNEME": {
"desc": "Temporal KG-RAG · 187 docs · 4 domains · 2020–2025",
"stack": "Sentence-BERT + Louvain + RRF",
"mrr": 0.85,
"status": "🟢 Submitted",
},
"Optimizer-Skin": {
"desc": "Late-fusion multimodal optimizer comparison · ISIC 2024",
"configs": ["Uniform Adam", "Decoupled LR", "Asymm SGD+NAG/Adam", "AdaGrad"],
"status": "🔵 In Progress",
},
}"Every green square is a step in gradient descent."
🇹🇷 Kayseri, Turkey — AGU · Research base
🇳🇱 Amsterdam, NL — ILA · 3 years Software engineering
🌐 10+ countries — EVS/ESC volunteering across Europe
Welcome to my writing space where engineering meets creativity.
--
✨ I actively publish on Medium about Flutter, and AI — turning research concepts into practical, scalable, and developer-friendly systems.
🔗 🌐 View All Articles on Medium
"Code is temporary, but architecture and ideas last."