Kodjo Jean DEGBEVI — Data Scientist, Analyst & AI Engineer.
I build ML systems end-to-end: raw data → pipeline → model → production API.
focus = [
"NLP for low-resource African languages (Ewé)",
"Remote sensing & geospatial AI (Sentinel-2, Google Earth Engine)",
"Supervised ML & deep learning (CNNs, fine-tuned Transformers)",
"API serving & MLOps (FastAPI, Docker, Gradio)",
"Data analysis & interpretability (EDA, SHAP, feature engineering)",
]ML & Deep Learning
NLP
Data & Vision
Serving & MLOps
Frontend & Tooling
|
Fine-tuning of NLLB-200-distilled-600M (Meta AI) for bidirectional Ewé ↔ French neural machine translation. Low-resource NLP for an under-resourced West African language — my mother tongue.
|
Full-stack ML application for credit risk prediction in financial inclusion. Scikit-learn model served via FastAPI; SHAP for interpretability; Next.js dashboard; one-command Docker Compose deployment.
|
||||
|
End-to-end AI system for land cover classification & deforestation monitoring in Togo — from data engineering (Sentinel-2 + Google Earth Engine) to model deployment.
|
Design and training of a CNN for handwritten digit recognition (MNIST). Full pipeline: preprocessing → architecture → training metrics → saved Keras model artefacts.
|
| Repo | What it is |
|---|---|
| perceptron-lab | Perceptron & 2-layer MLP from scratch with NumPy — weights, activations, backprop, gradient descent |
| ml-data-analysis-pipeline | EDA + ML modelling on Iris / Penguins / Titanic — KNN, Decision Tree, SVM |
| framework-ml | Academic ML workflow: data preprocessing, CatBoost / scikit-learn modelling, evaluation |
| tensorflow-pytorch-spacy-nltk-transformers | Notebooks covering TensorFlow, PyTorch, spaCy, NLTK for NLP & Computer Vision |
| MathNode | Math chatbot — LLM + SymPy for symbolic reasoning & interactive quiz generation (in progress) |
| quicknotes | Lightweight note-taking app built with Next.js |
| pythonInstaller_pyenv | Shell script to install & manage multiple Python versions via pyenv on Linux |
| streamer | Screen-casting from Fedora to TV / browser over a local network (VLC / OBS) |
