Imed Eddine Aouidane.

Python / pandas / NumPy / statsmodels / SQL / causal inference / forecasting

Completed at Djezzy, one of Algeria's leading mobile operators, on the DSIP4 Data Science Immersion Program (22 June – 30 July 2026). I owned the data and intelligence track behind a natural-language business assistant, working across 13 months of daily commercial KPIs — data catalogue, data quality framework, causal analysis, and the forecasting layer.

• Designed and applied a six-dimension data quality framework (completeness, accuracy, consistency, timeliness, uniqueness, validity) with severity tiering, encoded as automatic query-time guards so the assistant could not return unvalidated results.
• Established the lead–lag structure between commercial KPIs using cross-correlation, Granger causality and cointegration (ECM), separating genuine drivers from spurious correlation under HAC-robust inference.
• Built the forecasting and scenario service: 126 champion models selected from 245 candidates by rolling-origin backtesting against a seasonal benchmark, with prediction intervals published only where empirically calibrated and evidence-graded scenario simulation.

Described at a methodology level — Djezzy's commercial data, KPI definitions and findings stay confidential.

Python / PyTorch / ResNet-50 / OpenCV / Grad-CAM

Built a multi-label deep learning pipeline for cardiomegaly, pneumonia, and pleural effusion detection using the NIH Chest X-ray dataset with 112,000+ images and 14 disease labels.

Python / Telegram Bot / Claude API / Notion API

Designed a Telegram-based assistant for a cardiologist at CHU Mustapha Hospital, enabling natural-language access to structured patient records, medication notes, and clinical summaries.

R / ARDL / ECM / ARIMA / Econometrics

Modeled the causal relationship between AI adoption and NVIDIA revenue using quarterly financial data from 2015–2024, with ARDL bounds testing and error-correction modeling.

Python / SARIMA / XGBoost / LSTM / pandas

Compared statistical and ML forecasting methods on daily climate data to predict temperature trends and flag extreme heat events using engineered lag and interaction features.

PyTorch / TorchVision / PostgreSQL / Flask

Built a GPU-optimized Faster R-CNN inference pipeline with multi-threaded camera management, PostgreSQL event logging, and automated alert generation.

Python / PostgreSQL / Selenium / Power BI

Scraped and normalized 2,500+ product records from a cosmetics e-commerce competitor, enabling structured market positioning and competitor benchmarking.

Python / Streamlit / scikit-learn / seaborn

Developed a Streamlit application for bottom-hole pressure prediction with EDA, model performance reporting, and client-ready export workflows.

OpenAI Whisper / NLP / Python / FastAPI

Founded and built an AI meeting platform with real-time transcription, speaker diarization, summarization, secure authentication, and data workflows. Recognized as an innovative project by Algeria's Ministry of Knowledge Economy and Startups.

Brand Strategy / Logo Design / Packaging / Apparel

A separate creative venture showing visual direction, brand building, and product taste. It supports the portfolio without competing with the main ML narrative.