Sahnoun Bilel

– Designed an AI agent that generates store-specific audit checklists by leveraging each store’s audit history,

non-compliance patterns, and corrective actions.

– Built a multi-agent orchestration combining rule-based constraints with contextual signals (store profile, category,

seasonality, recent issues).

– Developed a RAG pipeline using CamemBERT embeddings, FAISS, MongoDB, and an Ollama-based LLM to

ground recommendations in internal documents and past audits.

– Integrated structured audit reports, templates, and field feedback into standardized, actionable checklist outputs.

– Implemented a feedback/learning loop with Stable-Baselines3 PPO and a custom OpenAI Gym environment to

improve checklist prioritization and relevance.

– Standardized store data and agent interfaces via the Model Context Protocol (MCP) to ensure interoperability and

easier integration.

Impact: Improved audit compliance across stores and increased the consistency and actionability of generated checklists.

– Extracted and parsed high-frequency time-series from MDG servers using robust regex-based pipelines and

automated checks.

– Built a reproducible preprocessing workflow (cleaning, encoding, scaling, temporal alignment) with leakage-safe,

time-based splits.

– Implemented and compared deep forecasting models (RNN, LSTM, DeepLOB) with walk-forward evaluation and

ablation testing.

– Improved performance through systematic hyperparameter tuning (sequence length, hidden size, dropout, learning

rate) and model calibration.

Impact: Achieved 60% accuracy on short-term trend prediction, improving the reliability of analyst monitoring.

– Prepared data through cleaning, encoding, and scaling to ensure optimal integration with machine learning models. – Built a complete pipeline using ColumnTransformer and Pipeline to streamline data transformations and model predictions. – Compared multiple classification models including Logistic Regression, Decision Tree, Random Forest, and XGBoost, using metrics such as accuracy, precision, and recall. – Developed a user interface with Tkinter enabling real-time prediction of employee turnover risk.