profile = {
"name" : "Samuel Hounsou",
"alias" : "hounsoubenny-cyber",
"school" : "IFRI Cotonou — Information Systems, L1",
"focus" : ["AI Engineering", "Cybersecurity", "MLOps", "FullStack Developper"],
"currently" : "Building ShieldAI V2 — ML-powered web vulnerability scanner",
"belief" : "Understanding internals beats using abstractions.",
"goal" : "Anthropic · Research Engineer",
}I don't copy tutorials — I build real systems. At 17, I've shipped a phishing detector at 99.3% accuracy, a hybrid IDS with <100ms latency, and a full-stack document classifier with OCR, JWT auth, and Fernet encryption. Every model, every pipeline, every line of backend — written from scratch.
|
URL phishing detector trained on 247K URLs. Stacking ensemble (XGBoost + HistGBT + RF) with Bayesian optimization and smart whitelist caching.
|
Hybrid network intrusion detection combining LSTM autoencoders, CNN, Isolation Forest, and LOF. Real-time multi-interface capture, Optuna-tuned, <100ms per packet.
|
|
Full-stack document classifier — PDF, image, text input via OCR. FastAPI + React, JWT auth, Fernet encryption, rate limiting, async file processing.
|
Auto-ML library that detects binary / multiclass / multilabel / regression from a single config file. Trains stacking ensembles with Bayesian optimization automatically.
|
📡 NEXUSFully offline RAG assistant built in 24h at a hackathon. Three search modes (semantic, keyword, hybrid), no internet required, ~500ms response time.
|
AI / ML
Backend / Security
Frontend
▸ Phase 5/9 of self-directed ML curriculum [NLP / LLMs / HuggingFace / RAG / Fine-tuning]
▸ ShieldAI V2 [Connecting ML pipeline + site mirroring]
▸ AetherFit / HELIOS [CPU-native LoRA + INT4 quantization research]
▸ Learning C++ [For low-level ML kernels]