AI & Full-Stack Developer passionate about building intelligent systems and scalable web applications — from beautiful frontends to production-grade ML pipelines.

• 🤖 Passionate about Deep Learning, Computer Vision, and NLP
• 🌍 Based in Morocco | Open to remote opportunities
• 🎓 Bootcamp graduate in AI Development from Simplon Maghreb (Sep 2024 – Mar 2026)
• 🔭 Currently exploring LLMOps, MLflow pipelines, and RAG architectures

Scrapy · PySpark · HuggingFace · ChromaDB · LangChain · Gemini · FastAPI · Next.js 15 · Airflow · MLflow · Prometheus/Grafana · Docker

Production-grade real estate intelligence platform built end-to-end. Features a stealth scraper with TLS anti-detection and behavioral rotation, a Bronze→Silver→Gold medallion data architecture on PostgreSQL/Supabase, and a Gradient Boosting model for price estimation with an undervaluation score. Includes HDBSCAN/KMeans clustering for opportunity detection, a contextual RAG chatbot (LangChain + ChromaDB + Gemini), a JWT-secured FastAPI backend with semantic search, and a Next.js 15 frontend — all orchestrated via Airflow and monitored with MLflow + Prometheus/Grafana.

LangChain · ChromaDB · HuggingFace · Gemini · MLflow · Kubernetes · PostgreSQL

IT support assistant powered by Retrieval-Augmented Generation. Ingests PDF documentation via PyPDFLoader, generates HuggingFace embeddings indexed in ChromaDB, and uses a controlled RetrievalQA chain with Gemini as the LLM. User questions are clustered with KMeans and stored in PostgreSQL. Experiments tracked with MLflow, deployed on Kubernetes via Minikube/Lens.

Airflow · PySpark · FastAPI · PostgreSQL · JWT · Docker

Big Data MLOps pipeline for urban delivery time prediction using NYC Taxi data. Airflow DAGs orchestrate a Bronze→Silver→ML workflow, PySpark handles distributed feature engineering, and a serialized regression model is served via FastAPI with JWT security and advanced SQL analytics (CTEs via SQLAlchemy).

Binance API · PySpark · Airflow · FastAPI · JWT · Docker

End-to-end fintech pipeline for predicting Bitcoin price 10 periods ahead. Ingests OHLC data from Binance into a Bronze zone, applies distributed feature engineering (returns, moving averages, taker ratio) with PySpark, trains a time-series regression model, and exposes predictions through a secure JWT FastAPI. Full Airflow orchestration with Docker Compose deployment.

Azure SQL · Azure AI Language · Terraform · FastAPI · Streamlit · OpenTelemetry · GitHub Actions

Cloud-native HR intelligence platform on Azure. Infrastructure provisioned via Terraform; Azure AI Language NER extracts skills from job postings automatically; salary estimation via a regression model. Features a FastAPI backend + Streamlit frontend, full Docker containerization, distributed tracing with OpenTelemetry/Jaeger, and a CI/CD pipeline (Flake8, Pytest, Docker build) via GitHub Actions.

Scikit-learn · FastAPI · Gemini · Next.js · PostgreSQL · JWT · Docker

Full-stack HR analytics app that predicts employee attrition. Trained LogisticRegression and RandomForest models with GridSearchCV, integrated into a JWT-secured FastAPI with PostgreSQL history logging. A /generate-retention-plan endpoint connects to Gemini to produce AI-generated retention strategies. Deployed via Docker Compose with a Next.js frontend.

BART · HuggingFace · Gemini · FastAPI · PostgreSQL · JWT

News analysis platform combining Zero-Shot classification (BART via HuggingFace) with Gemini for contextual summarization and tone detection. JWT-secured FastAPI backend with /analyze, /register, /login endpoints, PostgreSQL for user and log management, and full unit tests with mocked HuggingFace + Gemini calls.

BERT · HuggingFace · FastAPI · Next.js · JWT · Docker Compose

Multilingual NLP platform with sentiment analysis (nlptown BERT) and FR↔EN translation (Helsinki-NLP). Two JWT-secured FastAPI services expose /login, /predict, and /translate. Frontend built in Next.js with client-side JWT handling. Fully dockerized with Docker Compose.

TensorFlow/Keras · LSTM · MinMaxScaler · Python

Multivariate LSTM model for smart grid energy consumption forecasting. Implements chronological sorting, MinMaxScaler normalization, 24h windowing sequences, and a TensorFlow/Keras LSTM architecture. Model performance evaluated via loss/val_loss curves with real vs. predicted visualization.

Scikit-learn · FastAPI · SQLite · SQLAlchemy · Pydantic · pytest

REST API for cardiovascular risk prediction. Features a serialized Scikit-learn pipeline (joblib), CRUD endpoints + /predict_risk route, SQLite persistence via SQLAlchemy, Pydantic validation, Swagger documentation, and pytest + TestClient unit tests.

CNN · TensorFlow/Keras · OpenCV · FastAPI · PostgreSQL

Real-time emotion detection pipeline. Custom CNN (Conv2D, MaxPooling, Dropout) trained on augmented image data (rotation, zoom, flip), face detection with OpenCV Haar Cascade, and a detect_and_predict.py inference script. Served via FastAPI with PostgreSQL for prediction history.

TensorFlow/Keras · HuggingFace · DNN · EarlyStopping

Deep neural network achieving >80% accuracy on Fashion MNIST. Includes data normalization, EarlyStopping callbacks, and a benchmark comparison against a HuggingFace hosted model.

Scikit-learn · RandomForest · SVR · GridSearchCV · GitHub Actions

ML regression pipeline for logistics ETA prediction. Scikit-learn pipeline with StandardScaler, OneHotEncoder, and SelectKBest. Hyperparameter tuning via GridSearchCV (RandomForestRegressor & SVR), evaluated with MAE and R², and automated CI via GitHub Actions.

Scikit-learn · LogisticRegression · RandomForest · pytest · GitHub

Supervised ML pipeline for telecom churn prediction. Full EDA with Seaborn/Matplotlib, preprocessing pipeline (encoding + normalization), model comparison (LogisticRegression vs RandomForest), evaluation via Accuracy, F1-score, ROC and PR curves, unit tests with pytest, and GitHub versioning.

HuggingFace · ChromaDB · Scikit-learn · Evidently AI · Prometheus · Grafana · Kubernetes · GitHub Actions

Production MLOps pipeline for IT support ticket classification. NLP preprocessing (cleaning, tokenization, stopword removal), HuggingFace embeddings in ChromaDB, Scikit-learn classifier (F1/Recall), data & prediction drift monitoring with Evidently AI, infrastructure monitoring (CPU/RAM/containers) via Prometheus + Grafana, Kubernetes batch deployment (Job/CronJob), and CI/CD via GitHub Actions.