Ahmed Lachtar

Modeling Risk Frequency and average cost for dierent car and home in surance guarantees. Automation of pipelines through MLOps and leading

a car data enrichment project.

• Modeling of frequency, average cost and geographical rating for

non responsible damages.

• Helped automating pipelines for data processing, variable selection, modeling and model performance analysis for risk estimation using

the MLOps paradigm.

• Launching a vehicule data enrichment initiative through internal

data (underwriting, claims...) and external data (vehicule pricing, repair estimations...)

• Calibrating the Risk Premium through annual budget estimations from reserving.

• Analysis of Cost/Revenu for multiple client segments used for stake

holder decision making.

MLFlow, MLOps, Kedro Python, Databricks,Azure Microservices

On car insurance and home insurance products, developed fraud detec

tion algorithms and deployed them in production. Estimated savings of 5% on GWP per year. The algorithms reject automatically some con

tracts before subscription and creates alerts for less risky cases that are

handled by insurance agents.

• Created a dozen of alerts for dierent contract events (underwrit ing, car change or home move and portfolio)

• Automatic validation of customer details via external databases

and generating alerts accordingly (vehicle characteristics, clients with bad insurance history...)

• Creating a pipeline to automatically generate an alert if the cus

tomer is considered as risky and block him before subscription in some cases.

• Helped develop a guide for manual fraud alert processing used by insurance agents.

• Detecting and preventing organized group fraud through a graph based approach and implementing measures to prevent them from

underwriting.

• Using telematics data to detect fraud on declared annual mileage for clients that do not have the telematics product.

• Analysis and debugging of a pricing algorithm for clients with more

than one contract.

• Creating dashboards for fraud detection monitoring and other busi

ness needs.

• Maintaining an API for oer recommendations.

Python, Databricks,Azure Microservices, NetworkX, SQLAnalytics

A payment startup/ A french banking group: AML/CFT

This startup is a leader in marketplace payment services in Europe and is a client of a french Bank. We worked with a team of attornies from EY to solve compliance challenges related to money laundering and financing

terrorism. Tasks:

• Analyze adeaquacy of the warnings that the tools produces with the

regularory obligation of the startup and the bank

• Detect warnings related to data quality problems and provide quick and

long term solutions for these problems

• Verify the adeaquacy of the rules in the AML tool with the special type

of accounts used by the startup and suggest new rules based on AML

obligations for cantonment accounts

• Suggesting new rules based on a classification model from annotated data