Mahmoud Trigui

Mahmoud TRIGUI « Data Scientist » Street El Ain Km 6, Environment Boulevard 3042 Sfax – TUNISIA Mobile phone TN : (- E-mail-Self-tape https://youtu.be/OUljBuMxoXU LinkedIn www.linkedin.com/in/mahmoudtrigui Skill IQ https://app.pluralsight.com/profile/mahmoud-trigui Acclaim www.youracclaim.com/users/mahmoud-trigui/badges Zindi https://zindi.africa/users/Mahmoud_Trigui/competitions Personality https://www.16personalities.com/profiles/46e44008a0245 Titled with a statistics and data analysis engineering degree with a solid problem solving thanks to my mathematics-physics background, having +9 years of experience, passionate about anything related to machine learning, loving dealing with structured data, exploring data-mining insights for our daily issues tabular data, finding and creating new features through detailed analysis and approaching projects with different perspective. I use primarily R language but I'm always open to add more  Key Words Analysis, ANOVA Analysis, Business, Clustering Relationship, Construction, Churn, CVM, CRISP-DM ,Data storytelling, Dataiku, Decision rules, Dimensionality Reduction, Embedding, Feature engineering, Forecasting, Geospace Analysis, GPT-4, Hypothesis, IBM Watson, Image conversion, LLM, Methodologies, MLForecast, Network analysis, NLP, Oracle DB, Presentation, Profiling, Pydantic, R Markdown, SAS-GUIDE, SAS E-MINER, Sampling Techniques, Shiny App, Segmentation, Spatial Data, Transformations, Time Series Decomposition, Tutoring, VBA Contents Data Scientist @ Sofrecom Tunisia (part of Orange Group).......................................................................................... 3 PDF Enfo Extraction with LLM ........................................................................................................................................................... 3 Time Series Forecasting for Business Demand ................................................................................................................................ 3 CoFinancing Prediction for FTTH Deployment ............................................................................................................................. 4 Data Scientist @ Kiota Intelligence ................................................................................................................................. 4 Shiny App .................................................................................................................................................................................................... 5 RMarkdown Report................................................................................................................................................................................. 5 Exploratory Data Analysis ..................................................................................................................................................................... 7 Matching Script ......................................................................................................................................................................................... 8 Clustering Relationship .......................................................................................................................................................................... 8 Data Scientist/Analyst @ Tunisie Telecom ....................................................................................................................... 9 Client Segmentation ............................................................................................................................................................................... 10 Multi-Sim Detection .............................................................................................................................................................................. 13 Data-Mining Process .............................................................................................................................................................................. 14 Family Communities .............................................................................................................................................................................. 15 Ad-hoc Analysis / Reports ................................................................................................................................................................... 15 Data Science Projects @ Freelance .................................................................................................................................. 18 Facebook Page Analysis ........................................................................................................................................................................ 18 Vehicle Weight Estimation.................................................................................................................................................................. 18 Data Science Tutor.................................................................................................................................................................................. 19 Research Statistician @ LAAS-CNRS ............................................................................................................................. 22 Alarm Automatic management ...........................................................................................................................................................22 Machine Learning Competitor @ Zindi ............................................................................................................................. 24 Spatial Data .............................................................................................................................................................................................. 24 Various ........................................................................................................................................................................................................26 Assessments, Courses & Skills .......................................................................................................................................... 28 Pluralsight .................................................................................................................................................................................................28 LinkedIn .....................................................................................................................................................................................................28 Acclaim .......................................................................................................................................................................................................29 DataCamp ..................................................................................................................................................................................................29 Coursera .....................................................................................................................................................................................................30 edX ...............................................................................................................................................................................................................30 Stanford Online ....................................................................................................................................................................................... 31 MIT ProX, Udacity & DataQuest ...................................................................................................................................................... 32 + Personality ...................................................................................................................................................................... 34 Data Scientist @ Sofrecom Tunisia (part of Orange Group) PDF Info Extraction with LLM We developed a PDF Audit Information Extraction System using GPT-4 that automatically processes multipage audit reports. The system converts PDFs to images rather than text because some information are already on logos and graphics, it also search a specific items in a pre-defined reference dictionary. Built with a modular Python architecture, it features custom Pydantic validation models, optimized image conversion, and Base64 encoding for LLM compatibility. The solution generates structured JSON outputs while maintaining data integrity through comprehensive validation mechanisms, significantly reducing manual processing time for audit professionals. Time Series Forecasting for Business Demand The goal is to generate 12-month & 53-week demand predictions at multiple business levels (department & offer). The system addresses the challenge of limited historical data through advanced feature engineering, incorporating holiday calendars, weekend patterns, and seasonal decomposition techniques. My implementation leverages R's specialized forecasting packages (feasts, fable) to create an ensemble of models including ARIMA, ETS, Holt-Winters, TBATS, Prophet and Deep learning. The solution achieves approximately 5% of error. Key technical components include time series decomposition (STL, AutoSTR) to separate trend and seasonality patterns. CoFinancing Prediction for FTTH Deployment The goal is to predict co-financing rates for FTTH. The project involved migrating Python scripts to production-grade Dataiku workflows, prioritizing SQL and native components for performance optimization, and designing an efficient Analytics Base Table creation process. For model enhancement, we validated MLForecast function effectiveness, tested additional predictive variables, and evaluated alternative modeling approaches including other time series models and transforming the problem into a regression task. The part of feature engineering did improved the model, including creating interactions terms and doing the necessary transformations to have similar distributions across different period time. Data Scientist @ Kiota Intelligence My main work was to create the desired output for our endpoints with plumber, all type of task in data science with R included. All data were investment and funding rounds related. Shiny App The goal was to provide a tool that allow clients to have statistics and graphs about their data-entry based on many filters. R Markdown Report Generate a complete psychometric report about entrepreneurs and teams EDA The goal is to have a descriptive statistics and a test hypotheses between different stages in proportion. Matching Script The goal of this project was to connect the right investors to the right entrepreneurs based on many filters. Clustering Relationship The goal of this project was to find any potential relation between different variables and to connect their centroids to see the evolution over time. G Data Scientist/Analyst @ Tunisia Telecom Group I had the opportunity to work with big amount of data, millions of rows including hundreds of variables describing clients through different source and tables. A large amount of time for the data-mining project was dedicated to the ABT construction via:  Data reliability-check  Data gathering  Data cleaning So, in order to accomplish these tasks, I need to have a provident coding in SQL for Oracle (Toad), in addition to the basic programming of SAS and its functionalities in SAS Guide. Client Segmentation The business needed a new segmentation for our biggest part, the prepaid client base, to use it in theirs targeting campaigns, in the CVM and in various reports requests. The initial phase was the data preparation along with the schedule and the specification & mapping document. Phase 2 was the important one: trying different segmentation methods, different methodologies (one client per row or one client per period). Work done with SAS E-MINER. Primordial step to validate our clustering result is the detailed profiling, which will be affected by the business strategies and goals of the department. Phase 3 was for the supervised task: once the clusters are finally reviewed and validated, we build a model to assign every customer of each new month to its segment. Here we took a different approach in order to make it easier for the non-technical audience and for the directors, not by applying simply a classification model but to try to set easy decision rules extracted from the previous decision tree-based one, and then with the help of Tableau or Excel, we fit the final groups. Multi-Sim Detection We had to do several analytics to build initial hypothesis for our model since we do not have any record of multi-sim users in our databases, the idea was to proceed with clustering but with different approach (the goal to identify even small groups or who had a very practical behavior). With a detailed profiling, we were able to distinguish the most suspicious group, validate this by doing a callcenter campaign, then build a predictive model on top of that data (with sampling techniques). CRISP-DM I was integrated partially in the churn project among other ones, but every data-mining process is like: Family Communities Our Community Link Analysis system provides us with only the groups and their roles, but with no further information about the relations between clients of each community, thus the idea to find which community are representing one family and living under the same roof, the key data here was the Cell_ID tables, along with a hard work on Toad for feature engineering (used the ones generated by the CLA such as: betweenness, hub score ..), to finally use the appropriate network analysis package from R for the final touch. Ad-hoc Analysis / Reports Various demand from different department came to us to solve, to investigate and to find a patterns from our sources in the final goal to make the data storytelling. Understand the business behind the demand, create the right SQL request, organize the extracted data with Excel, and at some points go further with PowerPoint for presentations. VBA Data Science Projects @ Freelance I had the opportunity to work on various topics and connect to different teams with the agile methodology. Facebook Page Analysis The main idea was to do an overall analysis for one particular page which we have its credentials, to look at its numbers insights and to do some text mining in order to get sentiment analysis. Vehicle Weight Estimation To optimize its logistic line, a company want to predict the mass of vehicles on a connected weighing system with a maximum error less or equal to 5%. Data Tutor Tutor I had my chance to be a tutor and introduce a course about Machine Learning: algorithms and examples for a class of 5 persons. I also had another training session where I animated about the use of machine learning with IBM Watson for a class of 20 persons. Research Statistician @ LAAS-CNRS Alarm Automatic management The idea is to develop an automatic model to know from the owner's habits, if it's the owner who tries to enter or it's an intrusion based on the owner behavior, and then decide to launch the alarm or to deactivate it automatically. Machine Learning Competitor I have been competing for over 18 months & over 14 challenges and I’m not intending to stop any way for a simple reason: I just keep learning new things every day, new algorithm, new way of thinking, new approach, new trick, new package, new optimization. I gained so much knowledge through this period, because you need to be always up to date to latest news for the data science community or you will find yourself left behind and just repeating what you have gained! Spatial Data OSM & Roads Shape & Raster files, Projections Geospace Analysis Various Clustering Dimensionality Reduction ANOVA Analysis Assessments, Courses & Skills Pluralsight LinkedIn Acclaim DataCamp Coursera edX Stanford Online MIT ProX, Udacity & DataQuest English Certificate + Personality