Public research from an international open-source research lab: decision models, local AI agents, deepfake detection, GGUF builds and datasets.
Mert Kaya PRO
mertkayacs
AI & ML interests
Sr. AI Engineer, MSc & Sr. Fullstack Engineer
Gen AI / XAI / CV / LLMs / Agentic Workflows
TEDU MSc / TOBB ETU CS
Recent Activity
updated a Space about 6 hours ago
mertkayacs/postkitty-api published a Space about 6 hours ago
mertkayacs/postkitty-api updated a dataset 4 days ago
mertkayacs/emberwick-videosOrganizations
Run 2B and 4B language models locally: decision models with the JevAlt server and tool-use agents with Ollama or llama.cpp.
EfficientNet-B4 deepfake detectors with Grad-CAM facial-region analysis, FaceForensics++ results and research limits. Non-commercial weights.
-
mertkayacs/xdfdet
Video Classification • Updated • 1.79k • 1 -
FaceForensics++: Learning to Detect Manipulated Facial Images
Paper • 1901.08971 • Published • 1 -
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Paper • 1905.11946 • Published • 4 -
Improved Regularization of Convolutional Neural Networks with Cutout
Paper • 1708.04552 • Published
Decision-model evaluation and training data, executed AI-agent trajectories and village simulation videos. Read each card for splits and licenses.
Small 4B LLMs for text classification and decisions in English, Turkish and German, with probabilities, GGUF builds, demos and evaluation data.
A 2B tool-use model for local tables, notes and tasks. Full weights, GGUF for Ollama and llama.cpp, and executed training trajectories.
Karar-4B and Wähler-4B: small Turkish and German LLMs for text classification and decisions with probabilities. Full weights and GGUF builds.
Public research from an international open-source research lab: decision models, local AI agents, deepfake detection, GGUF builds and datasets.
Small 4B LLMs for text classification and decisions in English, Turkish and German, with probabilities, GGUF builds, demos and evaluation data.
Run 2B and 4B language models locally: decision models with the JevAlt server and tool-use agents with Ollama or llama.cpp.
A 2B tool-use model for local tables, notes and tasks. Full weights, GGUF for Ollama and llama.cpp, and executed training trajectories.
EfficientNet-B4 deepfake detectors with Grad-CAM facial-region analysis, FaceForensics++ results and research limits. Non-commercial weights.
-
mertkayacs/xdfdet
Video Classification • Updated • 1.79k • 1 -
FaceForensics++: Learning to Detect Manipulated Facial Images
Paper • 1901.08971 • Published • 1 -
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Paper • 1905.11946 • Published • 4 -
Improved Regularization of Convolutional Neural Networks with Cutout
Paper • 1708.04552 • Published
Karar-4B and Wähler-4B: small Turkish and German LLMs for text classification and decisions with probabilities. Full weights and GGUF builds.
Decision-model evaluation and training data, executed AI-agent trajectories and village simulation videos. Read each card for splits and licenses.