Tatakoto makes scientific research on artificial intelligence applied to medicine readable. Transformers for cancer detection, foundation models in imaging, clinical LLMs — we read the papers, explain them, and say what holds up and what doesn't.

The project

AI × health research has become the engine of 21st century medicine. Thousands of papers appear each year — on cancer detection in digital pathology, foundation models in imaging, LLMs in clinical reasoning, AI in drug discovery. Nearly all of it remains unreadable to anyone who isn't both a clinician and an ML engineer. Tatakoto sits between the raw publication and mainstream coverage that headlines press releases without checking the method.

The method

Each decryption follows a rigorous grid: model architecture, training data and its biases, honest comparator, metrics in ML terms AND clinical translation, generalization limits, model accessibility (code, weights, license), conflicts of interest, regulatory maturity. We name AI-health-specific failure modes: data leakage, shortcut learning, population bias, biased comparators.

Multilingual

Tatakoto publishes in French, English, Spanish, and Chinese.