This project was carried out as part of my apprenticeship at Socio Data Management.

General Description :
Creation of a modern, reusable survey questionnaire, including frontend and backend management, integration of advanced features, deployment, and maintenance.



Objective :
Initially managed by a colleague, the surveys were developed using ASP, a language no longer actively maintained. One major disadvantage was the inability to conduct interviews without an internet connection. This posed a significant problem, as certain locations might have limited or no connectivity, leading to wasted time, money, and missed interview opportunities. My goal was to migrate to more modern technologies to achieve better performance, advanced features, and long-term support.

Technical Details :
I developed the frontend using JavaScript, HTML, and CSS, enabling dynamic DOM creation and offering an interactive and responsive interface. On the backend, I used Flask to handle client interactions and MongoDB for NoSQL data management. The form model is automatically generated from an Excel sheet, making it easier to quickly create surveys for various studies.

Features :
The questionnaire supports both CAPI (Computer-Assisted Personal Interviewing) and CAWI (Computer-Assisted Web Interviewing).
CAPI Mode: Interviews are conducted in person using a tablet, allowing face-to-face data collection with the interviewer and interviewee.
CAWI Mode: Interviews are conducted online.
The questionnaire includes several advanced features, such as offline mode using Local Storage and PWA (Progressive Web App) technology, allowing users to complete surveys without an internet connection. Custom filters are supported, and an interview tracking system has been implemented. The order of question options can be randomized to ensure greater objectivity. Additionally, a quota management system has been added to limit responses per category.

Reusability and Deployment :
The questionnaire was designed to be easily reusable for other surveys, with adaptability to various contexts. The project was deployed in production for The Louvre, with more clients expected to follow. I continue to maintain the project, adding new features as needed.

Personal Reflections :
This project was not without its challenges. At times, I had to adjust my approach, such as when we decided to develop a web page instead of an extension to better track interviews. Sometimes, I encountered long obstacles, while at other times, the implementation went smoothly. With patience and a supportive team, I was able to build a robust and well-used tool. This project has been a rewarding experience, and I am proud of the outcome. It allowed me to develop skills in algorithmics, problem-solving, testing, and debugging. I grew in autonomy while also collaborating effectively with a team.

Timeframe : March - April 2025

Context : Project completed during the third year of the Data Science Bachelor's program at IUT de Paris – Rives de Seine, as part of a course on decision-making tools.

Description of the project :

City Fighting is an interactive web application that allows users to compare several French cities based on various aspects such as :

• Demographic data
• Job market
• Real estate and housing
• Weather data
• Points of interest
• Educational opportunities

Technical details :

The project uses APIs to collect and process real-time data, making it easy to organize and use. The data is then presented with Streamlit in an interactive and user-friendly interface, with dynamic charts and filters to explore the results.

Source :

• 🌍 Population : geo.api.gouv.fr
• 📊 General data, employment, housing, education : data.gouv.fr
• 💼 Job offers : data.gouv.fr
• ☁️ Weather : API OpenWeatherMap and meteostat.net
• 📍 Points of interest : API Overpass

Personal reflections :

This project required a lot of teamwork, especially through the use of Git for version control. I also learned how to collaborate effectively by working on and improving code written by others — a sometimes challenging but very valuable experience. It’s a project that could always be improved, but within the time we had, we did our best to deliver an application that is both intuitive and relevant.

General Description :
Application using a route calculator to suggest the most suitable journey based on personalized criteria such as speed, cost, or carbon footprint.


Technical Details :
Implemented with Streamlit using the Île-de-France Mobilités API, Geovelo, and Nominatim for address geocoding.

Source:
Île-de-France Mobilités API

Personal Reflections :
I would like to add a multimodal routing feature, such as bike+train, and merge the public transport section with the bike section on a single page. Although this project presented some challenges, working with API data was a very interesting experience.

General Description :
Applying software engineering best practices to the Titanic Survival Prediction project, using the collaborative tool Git, to predict the survival chances of Titanic passengers.

Technical Details:
Refactored the code into modular Python scripts, applied software engineering best practices (PEP 8, unit testing, documentation), collaborated via Git/GitHub, and automated the workflow with a CI/CD pipeline.

Source:
Kaggle - Titanic Tutorial

Personal Reflections :
A particularly enriching project in software engineering. Due to a lack of time, I was unable to integrate a Dockerfile.

General Description :
Application aimed at analyzing trends, optimizing content strategies, and identifying opportunities on the platform.


Technical Details :
Built with Shiny using the Plotly library for interactive charts.

Source:
Global YouTube Statistics 2023 - Kaggle

Personal Reflections :
Although YouTube's color is red, I thought it was better to diversify the colors used. The application remains clear and elegant.

General Description :
Application dedicated to investors to identify and select the most promising investment channels.


Technical Details :
Built with Shiny using the Plotly library for interactive charts.

Source:
Top Streamers on Twitch - Kaggle

Personal Reflections :
After lots of research and testing, I am satisfied with the result — simple and elegant.

With a group of three people, I conducted PCA and classifications, following an exploratory analysis, on football player data. The goal is to reduce the number of variables and to group the observations into clusters.
First, I performed the data cleaning. To conduct a good analysis of the players over the season, I chose to retain only those who participated significantly. Specifically, this means focusing on players who played more than 5 matches and had an average playing time of more than 45 minutes per match. To facilitate our analysis, I decided to consolidate the player positions into a single position per player: defender (DF), forward (FW), midfielder (MF), or goalkeeper (G).
Here is an example of PCA with axes 1 and 2:

The variables that contribute the most to axis 1 are SCA, PasAss, Shots, SoT, GCA, ToAtt, PPA, CPA. These variables are all related to offensive play, such as forwards. We can say that players on the right of the circle have an offensive play style, while players on the left of the circle have a less offensive play style.
The variables that contribute the most to axis 2 are Rec, Touches, PasAtt, Carries. These variables are strongly related to a style of play that involves a lot of ball control, such as midfielders. We can say that players at the top of the circle are those who control the ball a lot, while players at the bottom of the circle control the ball less.
However, it was not always easy to determine the number of clusters to keep.

I initially wanted to keep 6 clusters. But in reality, 3 clusters were enough, as we cut the tree where there is the largest jump.
I performed the processing and analyses using R. I used the explor package, which facilitates our study in the context of PCA. The output is an HTML file, generated with Rmarkdown.
This project greatly enhanced my teamwork skills and analytical capacity.

With a group of three people, I implemented a database. I was approached by the French e-commerce book company Nil.
The objective was to create a usable SQL database to understand user behavior on the platform. First, I started by designing the relational schema of our database, defining the name and function of each table, and then adding attributes to each table according to the requirements of the client.

I created the tables and the relationships using SQL queries in R with the DBI and RSQLite libraries. For the data implementation phase, I used existing web databases, client databases (name, first name, and phone number), and book databases (title, author, year of publication) from the site Kaggle. Then, I performed SQL queries to test the database and ensure everything was operational: checking data consistency, relationships, and date formats. Finally, I created a dashboard prototype based on the listed measures.
Special attention was needed for the date variable, as it had to follow a specific format (dd/mm/yyyy hh:mm).

This project greatly improved my teamwork, organization, and autonomy skills.

In a team of two, I analyzed a time series.
As part of a forecast study using data provided by the EIA (Energy Information Administration), the objective was to understand geothermal energy production from 2001 to the end of 2022 and to make predictions for the year 2023. The deliverable was a report.

The series is additive, with an irregular upward trend. These irregular variations caused by COVID-19 are explained by the trend curve and especially by the residuals. Outliers could explain unexpected variations. We observed a seasonal pattern that repeats each year.

I used three forecasting methods as indicated above. These three predictions are different from each other. However, they all have the same trend. After adjusting our predictions, we found that the Holt-Winters method is the best, as its prediction curve better fits the actual data compared to the other methods, and its mean squared error is the lowest.
This project allowed me to understand the specifics of time series data analysis.

Within the LAC distribution group (Leclerc, Auchan, Carrefour), our team (of 4 people) aims to structure data by implementing a database for decision-making purposes. Here, we are looking to monitor forecasting in the fruit and vegetable food sector. First, we had to create the conceptual and logical data model.

Then we cleaned and reformatted the data. We created the database using queries on Access. We also identified any anomalies. After calculating productivity and the rate of change, we were able to conclude whether or not to allocate headcount to the various sites.
I was the group leader. It wasn't always easy and straightforward. After all, we hadn't quite mastered Access at the time. It took a lot of trial and error to get to the end. There were a lot of queries to be made, which can quickly get us lost. We also have other projects at the same time, which doesn't leave us much time for rest and relaxation. But, thinking about the conceptual data model, implementing the database, and finding an answer to the problem based on this data are interesting and rewarding tasks. Because it prepares us for the eventual missions we may encounter in the professional world.

With my group of 4 people, I thought about a problem based on the data made available to me. This study is based on data from a database provided by the Observational Study in Clinical Epidemiology of 5 US hospitals that applied the RHC method. It concerns 5,735 patients over 18 years of age hospitalized in intensive care between June 1989 and January 1994. RHC, or right heart catheterization, consists in placing a Swan Ganz catheter to resuscitate a patient in serious condition. I wanted to see if this treatment, applied to intensive care patients, was beneficial, if it could save the most people, and more specifically to see if the RHC method was relevant and necessary in certain situations.
The goal was to produce a technical report, a statistical report, a project management report, and a final presentation of the results with slides. The technical report contains the approach I took on R to clean and sort our data. Indeed, it was necessary to understand and assimilate the data. After cleaning, my observations went from 5735 patients to 5508 patients. For example, I removed observations that had outliers on a variable. I decided to delete a variable that had a lot of missing values. I also grouped variables into one for easier manipulation later on, etc.
Then I made graphs and analyses where I added it to the statistical report. And I presented our results on Powerpoint during the oral presentation. Throughout the project, I drew up a project management report, which consisted in planning the project with a Gantt chart, organizing it, and seeing whether the expected time and resources had been respected.
I used R software for data processing and analysis. This project taught me a lot about teamwork, organization and autonomy.