Skip to content
View ayayoussfiii's full-sized avatar
5 followers · 11 following

Block or report ayayoussfiii

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ayayoussfiii/README.md 
 █████╗ ██╗   ██╗ █████╗     ██╗   ██╗ ██████╗ ██╗   ██╗███████╗███████╗███████╗██╗
██╔══██╗╚██╗ ██╔╝██╔══██╗    ╚██╗ ██╔╝██╔═══██╗██║   ██║██╔════╝██╔════╝██╔════╝██║
███████║ ╚████╔╝ ███████║     ╚████╔╝ ██║   ██║██║   ██║███████╗███████╗█████╗  ██║
██╔══██║  ╚██╔╝  ██╔══██║      ╚██╔╝  ██║   ██║██║   ██║╚════██║╚════██║██╔══╝  ██║
██║  ██║   ██║   ██║  ██║       ██║   ╚██████╔╝╚██████╔╝███████║███████║██║     ██║
╚═╝  ╚═╝   ╚═╝   ╚═╝  ╚═╝       ╚═╝    ╚═════╝  ╚═════╝ ╚══════╝╚══════╝╚═╝     ╚═╝

AI & Digital Trust Engineering · ENSA Fès, Morocco

LinkedIn GitHub Email Status Oracle OCI Views

class AyaYoussfi:
    school   = "ENSA Fès — Cycle Ingénieur · AI & Digital Trust (2024–2026)"
    focus    = ["Explainable AI", "RAG Systems", "AI Safety", "Real-Time ML"]
    stack    = ["Python", "PyTorch", "LangChain", "Kafka", "SHAP", "Docker", "React"]
    certs    = ["Oracle OCI AI (in progress)", "Cisco CTM", "Cisco NET", "UM6P IIoT"]
    looking  = ["Internship", "Research Collaboration", "Open Source"]
    motto    = "Build systems that explain themselves."

01   Featured Projects

🛡️ NetGuard IDS

Real-time host-based intrusion detection on the BETH Dataset (NeurIPS 2021) — 8M+ kernel syscall events from 23 real AWS honeypots. Isolation Forest + XGBoost + SHAP + RAG over MITRE ATT&CK.

# BETH — 8M+ events · 23 AWS honeypots
anomaly  = IsolationForest().fit(X_benign)
classify = XGBoost().fit(X_labelled)
explain  = shap.TreeExplainer(classify)
report   = rag.query(mitre_attack, alert)

⚡ Real-Time Jailbreak Detection

DistilBERT classifier streamed over Apache Pulsar. Hot-swap model watcher for zero-downtime updates. Adversarial prompt detection at sub-100ms latency.

# streaming adversarial prompt classifier
stream = PulsarConsumer(topic="prompts")
model  = AutoModel.load(hot_swap=True)
label  = model.classify(stream.next())
# < 100ms end-to-end latency

💳 Credit Card Fraud · HDBSCAN

30k clients · HDBSCAN segmentation → per-cluster Gradient Boosting → SMOTE oversampling → SHAP per-cluster explainability · Flask REST API · deployed on Vercel.

clusters  = HDBSCAN().fit_predict(X)
models    = {c: GBM().fit(X[c]) for c in clusters}
shap_vals = {c: TreeExplainer(m) for c, m in models}
api       = Flask().expose(models, shap_vals)

🏭 GuardianAI + ChurnAI

FastAPI AI security proxy with 3-layer middleware pipeline (auth, prompt filter, response audit). Paired with ChurnAI: XGBoost churn classifier + SHAP + LLM-generated personalised retention strategies.

app = FastAPI()
app.add_middleware(AuthGuard)
app.add_middleware(PromptFilter)
app.add_middleware(ResponseAudit)
# + structured pytest suite · polished README

📦 RAG Pipeline at ALTEN (internship)

Production RAG pipeline with Flask API. Vector retrieval + LLM-grounded answers over internal documentation.

embedder    = SentenceTransformer("all-MiniLM-L6-v2")
vectorstore = Chroma(embedder)
chain       = RetrievalQA(llm, vectorstore)
api         = Flask().expose(chain)

🦯 Smart Belt for Visually Impaired

Arduino + GPS + ultrasonic sensors · Real-time obstacle detection · Haptic feedback navigation at the edge · no cloud dependency.

while (true) {
  d = ultrasonic.read();
  if (d < THRESHOLD)
    haptic.pulse(map(d, 0, MAX, 255, 0));
  gps.track(location);
}

02   Stack

AI · ML · XAI

Python        ████████ 97%
PyTorch       ███████░ 82%
scikit-learn  ███████░ 85%
XGBoost       ███████░ 88%
SHAP          ███████░ 80%
TensorFlow    ██████░░ 75%

NLP · LLMs · RAG

LangChain     ███████░ 78%
HuggingFace   ███████░ 80%
ChromaDB      ██████░░ 72%
DistilBERT    ██████░░ 75%
RAG Systems   █████░░░ 70%
Transformers  ██████░░ 73%

Infra · Backend

Docker        ███████░ 80%
FastAPI       ███████░ 85%
Flask         ███████░ 85%
Kafka         ██████░░ 70%
PostgreSQL    ██████░░ 72%
Apache Pulsar █████░░░ 65%

Frontend · Data

React         ███████░ 72%
Tailwind      ███████░ 75%
Pandas/NumPy  ████████ 92%
SQL           ██████░░ 78%
Java          █████░░░ 65%
Bash          █████░░░ 68%

03   Certifications



Generative AI Professional
Oracle · In progress

Cyber Threat Management
Cisco

Network Technician
Cisco

Industrial IoT
UM6P

04   GitHub Activity

   

 

"Build systems that explain themselves."

Open to internships · research collaborations · open source · aya.youssfi@usmba.ac.ma

Popular repositories Loading

  1. ayayoussfiii ayayoussfiii Public

    Aya's README for GitHub

    1

  2. aya-portfolio aya-portfolio Public

    My personal portfolio website

    HTML

  3. agencecar agencecar Public

    AyaCar is a desktop-based vehicle rental management system developed in Java (Swing) with a Microsoft Access database. The application is designed to optimize the daily operations of a small or med…

    Java

  4. designpatternproject designpatternproject Public

    Le projet Tetris Galaxy Dream, une implémentation moderne du jeu classique Tetris développée dans le cadre du module Design Patterns

    Java

  5. CreditCardFraud-Project CreditCardFraud-Project Public

    Built an end-to-end credit card default prediction system on 30,000 clients. Used HDBSCAN for behavioral segmentation (3 clusters), Gradient Boosting per cluster with SMOTE for class imbalance, Nai…

    Python

  6. ChatBot--VoiceRAG-Assistant ChatBot--VoiceRAG-Assistant Public

    A voice-enabled AI assistant that answers questions about technical documents using a Retrieval-Augmented Generation (RAG) pipeline. The system indexes documents and retrieves relevant context to g…