Available for opportunities

Johannes
HOUNTON

AI Scientist from Benin. I don't just develop — I research, build, and push AI systems to their limits with whatever resources I have. Neural networks, NLP, Computer Vision, Audio Processing. Every challenge is a new frontier.

View Projects → Get in touch

johannes@ai-lab:~

$cat profile.json

{

"role": "AI Scientist",

"location": "Cotonou, Benin 🇧🇯",

"focus": [

"NLP", "Computer Vision",

"Audio Processing"

"languages": ["FR", "EN"],

"status": "building the future"

}

Expertise

Core Domains

🧠

Deep Learning

CNN, RNN, Transformers, Attention Mechanisms, Encoder-Decoder, CTC architectures. Low-resource GPU optimization.

🗣️

NLP & LLMs

RAG pipelines, Language Modeling, Translation, Sequence Models, Vector Spaces, Probabilistic Models.

👁️

Computer Vision

Segmentation (U-Net, Chan-Vese), Classification, Detection, Facial Recognition, Motion Tracking.

🎙️

Audio Processing

ASR, voice feature extraction, speech synthesis, librosa, soundfile. Deep Learning for audio.

⚙️

ML Engineering

TensorFlow, PyTorch, fairseq, scikit-learn. Domain adaptation, transfer learning, model optimization.

🚀

Deployment

Flask-based ML platforms, REST APIs, AI product integration. Low-resource environment adaptation.

Work

Selected Projects

01 Medical AI · Computer Vision

Stroke Segmentation

7-month research project for stroke detection on CT scan images. Double-level solution: U-Net with domain-adapted pretrained layers + Chan-Vese algorithm for hypodensity detection across 4 vascular brain areas.

02 NLP · Public Sector

ASP — Chatbot Public Service Benin

RAG-powered intelligent assistant answering citizen questions about public services in Benin. Built with agentic architecture for context-aware, accurate responses.

03 Audio · Speech

Audio Processing Suite

Multi-project suite for voice feature extraction and analysis. Mastering ASR and voice synthesis with deep attention mechanisms, CTC architectures, and GPU-optimized training in low-resource environments.

04 Low-Resource NLP · Africa

Fongbé Translation

Neural machine translation model built with fairseq to translate French into Fongbé, a low-resource West African language. Pioneering NLP for underrepresented languages.

05 Sign Language · Accessibility

AfriCanTalk

Sign language translation system for African languages. Bridging communication gaps by converting sign language into African spoken/written languages using computer vision.

06 Medical AI · Web Platform

Breast Cancer Classifier

Flask + TensorFlow web platform for breast cancer degree classification. Predictive diagnosis tool designed for early detection and medical support.

07 Computer Vision · Classification

DogVsCat — 37 Species

Advanced multi-class classification of 37 cat and dog species. Achieved 85%+ accuracy with loss below 0.5 through custom transfer learning strategies.

08 NLP · Insurance

Insu — Agentic RAG

Agentic RAG system combining regression analysis for medical insurance prediction with intelligent Q&A capabilities. Insurance cost forecasting with AI guidance.

09 Medical AI · Segmentation

SkinLesionFind

Image segmentation model for skin lesion detection and analysis. Deep learning approach to identify and delineate lesion boundaries for dermatological support.

10 Robotics · Object Detection

Trashbox Detection

Detection model identifying trashbox types and colors to guide a robot in sorting waste into the correct bins. Team project combining CV with robotic control logic.

11 Computer Vision

Facial Detection System

Face identification system comparing input images against a database using embedding similarity. End-to-end pipeline from detection to identity verification.

12 NLP · Hackathon

AudiBot — Indaba Ghana

RAG-powered sales assistant built during the Deep Learning Indaba Ghana hackathon with teammates. Fast-prototyped Q&A system for product and sales inquiries.

Johannes
HOUNTON

Core Domains

Selected Projects

Background

Let's connect

Building the
next generation
of AI.

JohannesHOUNTON

Core Domains

Selected Projects

Background

Let's connect

Building thenext generationof AI.

Johannes
HOUNTON

Building the
next generation
of AI.