+7%
Gini improvement vs GLM baseline
Pure-premium ranking on the evaluation set
Available now · International · 6 months
Mariem Fersi · Data Science Engineer
Make complex data useful for business.
I turn complex data into practical AI and risk tools, bringing engineering and actuarial thinking together to help teams make confident decisions.
AI & machine learningData engineeringActuarial & risk modeling
.png&w=3840&q=75)
Tunis · open to relocation
#2 / 27
Engineering class
A rare combination
AI × Actuarial
AI engineering · Actuarial thinking · Decision-ready data · Risk modeling
02 · Evidence
Better risk decisions need more than a prediction.
The business problem
How can insurers price and reserve risk with accuracy, calibrated uncertainty and an audit trail?
Traditional GLMs can miss complex patterns; black-box predictions can be difficult to trust. This project combines actuarial baselines, distributional learning and explainability.
A reproducible evaluation-to-service workflow: GLM baseline → CANN / NGBoost → conformal bounds → SHAP → FastAPI, MLflow and Docker.
Measured project outcomes
Deep Distributional Actuarial Modeling
91.9%
Conformal coverage
Empirical coverage measured on the evaluation set
0.815
Fraud AUC-ROC
Fraud-detection model on the evaluation set
Reported results come from the project's evaluation data; they are not production or client KPIs.
03 · Milestones
Proof built over time.
From quality operations to connected-vehicle analytics and actuarial AI—each step adds a new layer.
01
Jun – Aug 2023SOPAL
Quality & Environmental Intern
Process quality and compliance
- Monitored production and quality processes, standardizing documentation workflows.
- Supported environmental-compliance activities and audit-ready record keeping.
Chapter 01
02
Jun – Aug 2025Capgemini
Data Engineering & BI Intern
Data engineering for connected vehicles
- Built ETL pipelines in SSIS to consolidate connected-vehicle telemetry into modeled SQL Server schemas.
- Developed Power BI dashboards with DAX so analytics teams could act on the modeled data.
- Automated recurring data-preparation and quality-check workflows.
Chapter 02
03
Jul – Sep 2026Talan
AI / Actuarial Modeling Intern
Deep learning applied to insurance
- Designed an end-to-end AI architecture combining LLMs, machine learning and data pipelines for uncertainty-aware insurance pricing, reserving and fraud detection.
- Implemented distributional pricing experiments (GLM baseline, CANN, NGBoost) and conformal prediction to quantify reserving uncertainty.
- Evaluated GNN-based fraud detection with SHAP explainability, tracking experiments in MLflow and serving models with FastAPI/Docker.
Chapter 03
Two complementary degrees
ESPRIT
Engineering Degree in Data Science
2023 – 2027 · Tunisia
Le Mans University (IRA)
Master in Actuarial Science
2025 – 2027 · France
Academic performance
16.45 / 20
Ranked #2/27 in class
Second Engineering Year
16.31 / 20
Major of engineering class
First Engineering Year
04 / The work
Ideas, put to work.
Five projects. Real questions. A mix of AI, data and actuarial thinking.
0501 · Featured researchAI · Actuarial Science
Deep Distributional Actuarial Modeling
Pairs actuarial benchmarks with explainable AI to estimate risk—and how uncertain that estimate is.
PyTorchCANNNGBoostGaussian CopulaConformal Prediction
Explore the project
Watch demo · Actuarial AI Platform demo+
From prediction to decision
A complete modeling journey
01GLM baseline
02Distributional AI
03Calibrated risk
+7%
Gini improvement vs GLM baseline
91.9%
Conformal coverage
0.815
Fraud AUC-ROC
Evaluation results · not production or client KPIs
More from the portfolio
Different problems. Different tools.
02
MARKET · MACRO · SENTIMENT
Market signals
Specialist agents
Explainable view
AI Engineering · Quantitative Finance
Multi-Agent FX Intelligence
Brings market, macro and sentiment signals together through specialist agents, with a traceable rationale.
PythonLangGraphMulti-Agent
Explore project
Watch demo · Trady demo+
03
CLIMATE RISK · UNDERWRITING
Exposure data
Hazard & loss
Risk insight
Climate Risk · Reinsurance · AI
ImplementedPrototypeResearch / ProposedClimateGuard AI
Explores how climate and exposure data can help underwriters see catastrophe risk more clearly.
PythonAzureXGBoost
Explore project
Watch demo · ClimateGuard AI demo+
04
MORTALITY · LONG-TERM VALUE
Population data
Stochastic model
Annuity value
Actuarial Modeling
Mortality & Life Insurance
Uses stochastic mortality models to forecast changing lifespans and explore their impact on annuity value.
RStMoMoHuman Mortality Database
05
DOCUMENTS · AUDITABLE VALUE
PDF & Excel
OCR + validation
Model + report
Document AI · Insurance
In progressAI Asset Valuation & Insurance Risk
Turns valuation documents into checked, traceable inputs for insurance and asset calculations.
PythonPandasOpenPyXL
Proof beyond projects
Learning, made tangible.
05 · Your next great hire?
Have a complex problem? Let’s make it clear.
I’m available for a six-month international final-year internship. Let’s explore how AI, data engineering or risk analytics could move your team forward.
Back to the beginning