Stefan Strainovic

Led the development of a multimodal AI platform for non-invasive health screening in a bootstrapped startup environment.

• Designed and built a multimodal voice-based health screening model that converts audio recordings and transcripts into clinical scores using attention encoder architectures.
• Engineered end-to-end Python pipeline combining acoustic feature extraction (prosody, pauses, energy, spectral characteristics) with linguistic analysis and Transformer-based embeddings.
• Integrated automatic speech recognition, feature engineering, model training, and validation into production-ready system for continuous deployment and experiment tracking.

Developed and deployed production machine learning systems for credit risk assessment, automating financial workflows at scale.

• Built ensemble credit risk models (XGBoost, Random Forest) achieving >90th percentile accuracy and deployed production-grade scoring engine via Streamlit for credit decisions.
• Engineered automated transaction classification system using TF-IDF and semantic analysis, processing and labeling thousands of transactions with minimal manual intervention.
• Designed intelligent transaction prioritization algorithm leveraging k-means clustering to reduce manual labeling burden by 95%, from thousands to ~200 transactions per client per cycle.
• Prototyped CNN-based document parsing system for automated bank statement extraction from PDF images, enabling end-to-end document ingestion pipeline.
• Built comprehensive ML monitoring dashboards and automated validation pipelines ensuring model reliability and drift detection in production.

Designed and executed statistical analysis for radiation oncology research investigating immunological effects of brain irradiation therapy protocols.

• Conducted rigorous statistical analysis quantifying the differential impact of X-ray and proton brain irradiation on white blood cell (T-lymphocytes) subtypes (CD4+/CD8+/Treg populations).
• Developed multivariate regression models with proper diagnostic validation to identify significant correlations between irradiation protocols and lymphocyte concentrations.
• Produced publication-quality statistical reports and visualizations translating complex immunological findings for clinical and research teams.

• Statistical Theory and Methadology: Rigorous training in multivariate data analysis, classification, regression models, and time series analysis for forecasting and modeling stochastic processes.
• Implementation of statistical models and algorithms: Implementation using statistical software (Pythong, R and SAS), with a focus on Operational Research, Discrete Optimization and Machine Learning.
• Internships and Projects: Executed internships in statistical research (Oncology) and machine learning (Finance), as well as personal projects in Deep Learning applied to Medical Imaging.

• Core & Applied Mathematics: Gained a strong foundation in key mathematical fields including Algebra, Analysis, Differential Equations, and Optimization.
• Computational & Probabilistic Modeling: Developed proficiency in Probability, Statistics, and Numerical Analysis, complemented by skills in programming and algorithms.
• Project-Based Learning: Successfully completed multiple mathematical and pre-professional projects, applying theoretical knowledge to practical challenges and enhancing research and problem-solving ability.

Active member of the local ESN section board, participating in the organization of events for international students.

Organized events and activities for international students and represented the Caen section at the regional level.

Served as both a practicing artist and manager of a remote, international team of artists in the creation of digital assets for video games.