kheiriddine Dardouri

• Developed an on-premise solution using specialized OCR models for each document type (checks, invoices, bank statements, general ledger). • Designed a cloud-based architecture with AWS S3 and AWS Glue for scalable and distributed document processing Key Features: ✓ hybrid OCR system achieving 95% extraction accuracy ✓ LLM agents processing and converting outputs to queryable JSON ✓ A rules-based anomaly detector flagging 15+ discrepancy types ✓ Interactive dashboard with real-time OCR correction interface

Developed credit risk models to predict default probabilities for a commercial loan portfolio at Bank of Africa (BoA). Processed and staged client data for risk analysis, performing feature engineering to enhance model accuracy. Supported IFRS 9 compliance by implementing risk solutions in SAS and automating data quality checks. Key Features: ✓ IFRS 9-compliant staging logic (Stage 1, 2, 3 classification) ✓ SAS macros for automated data ingestion, transformation, and scoring ✓ Feature importance analysis to enhance explainability and regulator alignment ✓ End-to-end pipeline from raw data → risk score → staging classification → compliance reports via SAS Solution

• Designed an AI-based system to counter both narrowband and wideband reactive jammers using real-time learning and adaptation. • Modeled the interaction as a strategic game to dynamically adjust transmission parameters (e.g., frequency, power). • Integrated a transformer-based deep learning model to predict channel gains and optimize the algorithm’s performance.

• Built an AI-based system that detects and counters jamming attacks in real time. • Modeled the anti-jamming logic as a strategic game between users and attackers. • Dynamically adjusted frequency and signal power to maintain reliable communication.