Ivann Vasic_

About

Skills

Data Science Machine Learning Applied Mathematics Quantum Computing Programming Development Tools Language

Data Engineering & Cloud:Google Cloud Platform, BigQuery, PySpark, Apache Cassandra, Google Spanner.

Data Analysis & Manipulation: Python, NumPy, Pandas

Data Visualization: Matplotlib, Seaborn, Plotly, Power BI, Folium

Databases: PostgreSQL, MySQL, MongoDB, Firebase

Graph Data & Semantic Web: Neo4j, Graph Data Science, SPARQL, RDF Schema, Ontologies, OWL

Machine Learning: Scikit-learn, KNN , SVM, Gaussian Process Regression, Logistic Regression, Discriminant Analysis (LDA, QDA), Naive Bayes, Linear Regression

Unsupervised Learning: K-means, DBSCAN, PCA, FLDA

Large Language Models: GenAI, LangChain, Prompt Engineering, RAG, PEFT

Deep Learning: PyTorch, CNNs, Transformers, Computer Vision

Model Workflow & Optimization: Feature engineering, Cross-Validation, Model Selection, Transfer Learning, Fine-tuning, Data Augmentation, Multi-fidelity training.

Qubit Systems and models: Tensor product, Entanglement, Toffoli, Quantum Complexity

Quantum Cryptography: BB84 protocol, RSA

Quantum Algorithms: Deutsch, Deutsch–Jozsa, Grover, Shor, QFT

Quantum Machine Learning: Variational Quantum Circuits, Quantum Circuit Born Machine, Ansatz

Quantum Frameworks: PennyLane , Qiskit, MPQP

Linear Algebra: Vectors, Matrices, Eigenvalues/Eigenvectors, Projections

Probability: Distributions, Conditional Probability

Statistics: Descriptive Stats, Estimation, Confidence Intervals, Regression, Maximum Likelihood Estimation, Bayesian Decision Theory, Covariance, Correlation.

Loss functions & Optimizers: Kullback-Leibler divergence, Maximum Mean Discrepancy, Cross-Entropy, Adam, RMSprop, Cobyla, SPSA

Languages: Python, C, C++, Java, Kotlin, HTML, CSS, JavaScript, PHP, SQL

Frameworks:Symfony, Angular, Tkinter

Databases: PostgreSQL, MySQL, Firebase, MongoDB

Version Control: Git, GitHub, GitLab

IDEs: VS Code, JetBrains, Jupyter Notebooks, Android Studio, Google Colab

Containerization & Orchestration: Docker, Docker Compose

French - Native (C2)

English - Fluent (C1)

Chinese - Intermediate (HSK3)

German - Elementary (A2)

Timeline

• High School Degree (Baccalauréat)

  Lycée Polyvalent Louis Aragon

  September 2019 – July 2022 (3 years) Héricourt | France

  Specialization:

  • Mathematics
  • Physics and Chemistry
  • English

  Option:

  • Advanced Mathematics

  Result:

  • Awarded with high honors
  • 19/20 in mathematics specialization
  • 18/20 in physics and chemistry specialization
• Scientific Preparatory Cycle

  Université de Technologie de Belfort-Montbéliard

  September 2022 – January 2024 (1.5 years) Belfort | France

  Specialization:

  • Theoretical Mathematics
  • Physics
  • Chemistry
  • Computer Science
  • Engineering Fundamentals
• Operator Internship

  General Electric

  January 2023 – February 2023 (1 month) Belfort | France

  Assembly line for turbines on a base

  • Team Working
  • Management
• Innovation Hackathlon - Forvia

  Innovation Crunch Time

  March 2023 (1 week) Belfort | France

  Europe's biggest university technology challenge, 2023 edition

  • Machine learning project to design a car seat that can be adapted to all body types.
  • Working in a team of 10 students supervised by Forvia.
• Innovation Hackathlon Team Leader

  Innovation Crunch Time - GE Vernova

  March 2025 (1 week) Belfort | France

  Europe's biggest university technology challenge, 2025 edition

  • Leading a team of 10 students on an HR project to optimize skills management.
  • Project supervised by General Electric Vernova.
• Scientific Preparatory Cycle (Exchange Semester)

  Shanghai University (上海大学)

  February 2024 – June 2024 (5 months) Shanghai | China

  • Focus on computer science theory, Data Science, applied mathematics and chinese language.
  • Entrepreunership based group projects with international students.
  • Data Science project to optimize green spaces repartition in the megacity of Shanghai.
• Innovation Bootcamp

  Université de Technologie Sino Européenne de Shanghai

  June 2024 – July 2024 (1 month) Shanghai | China

  • One month of intensive work in an intercultural team, presented to an experienced jury.
  • Development of a startup project for a banking system in disadvantaged countries.
  • Entrepreunership, WEB Development, Team Work, Innovation
• Master degree in Engineering, Data Science

  Université de Technologie de Belfort-Montbéliard

  September 2024 – June 2027 (3 years) Belfort | France

  Three-year engineering curriculum combining advanced coursework, research projects, and industrial internships.

  • WEB Development
  • Data Science
  • Applied Mathematics
  • Quantum Computing
  • Machine Learning
• AI Engineer Internship, R&D

  Michelin

  September 2025 – February 2026 (6 months) Clermont-Ferrand | France

  Developed a statistical framework benchmarking Finite Element tools vs. AI models, leveraging Unsupervised Learning and Explainable AI to diagnose biases and drive the improvement of predictive performance.

  • Data Science
  • Machine Learning
  • Statistics
  • Applied Mathematics
  • Python
• Innovation Hackathlon Team Leader

  Innovation Crunch Time - HNFC

  March 2026 (1 week) Belfort | France

  Europe's biggest university technology challenge, 2026 edition

  • 5 days innovation challenge focused on Design Thinking
  • 1600 multidisciplinary students forming teams of 10
  • Leading a team of students under the supervision of Hopital Nord Franche Comté.
  • Developed a low-cost syringe heating solution using CATIA modeling and iterative prototyping.

Projects

Personal / Professional Projects

Physiovapp - Daily Health Tracker

A modern web app to track daily habits, health routines, and personal progress with real-time data and interactive charts.

Next.js, React, TypeScript, Firebase, Firestore, Recharts, Tailwind CSS

View Details

Team size: 1

Date: 2025 - 2026

Type: Personal Project

Physiovapp is a full-stack-ready wellness tracker focused on practical daily consistency and clear data visualization, that I use myself to improve productivity and health habits:

• Built a responsive web interface with Next.js and React, designed for both mobile and desktop usage.
• Implemented real-time daily logging with Firebase Authentication and Firestore for user-scoped data.
• Developed client-side analytics using Recharts (nutrition, sleep, activities, routines) with dynamic scales and multi-series charts.
• Designed routine tracking logic with morning/evening states to improve habit accuracy.
• Optimized architecture for fully free deployment on Firebase Hosting without paid backend runtime dependencies.

Code Website Link

HSK Trainer - Chinese Language Learning Platform

Web app for preparing HSK exams, combining FSRS with LLM-powered written-expression grading.

FastAPI, Next.js 16, PostgreSQL, Anthropic API, Python, TypeScript, Tailwind CSS

View Details

Team size: 1

Date: 2025 - 2026

Type: Personal Project

HSK Trainer is a production-grade language-learning platform covering HSK levels 1–3 (~600 words), built as both a daily study tool and an engineering portfolio piece:

• Implemented an FSRS spaced-repetition scheduler that models per-card stability and difficulty, scheduling each review just before predicted forgetting — outperforming classic SM-2 (Anki).
• Integrated Anthropic API (streaming SSE) to grade free-form written Chinese, returning a structured correction: score, corrected version, pinyin, translation, typed mistakes, and a next-step recommendation.
• Designed a feedback loop that automatically re-queues vocabulary flagged as misused by llm back into the FSRS review deck.
• Built an optimistic quiz UI (Next.js + TanStack Query) that prefetches the session and advances client-side on rating, reducing perceived latency from 2–5 s to under 50 ms.
• Deployed the full stack at near-zero cost: Vercel (frontend), Render free tier (FastAPI backend), Neon Postgres — with a keep-warm cron and infrastructure-as-code via render.yaml.

Code Report Website Link

School Projects

Bank Term Deposit Prediction - ML Binary Classification

Machine learning study on a Kaggle Playground dataset, benchmarking six classifiers and tackling severe class imbalance and data leakage.

Python, Pandas, Scikit-learn, XGBoost, LaTeX

View Details

School: UTBM

Team size: 3

Date: 2026

Course: Data Science & Machine Learning

A full supervised binary classification pipeline predicting whether a client subscribes to a term deposit after a marketing campaign:

• Performed in-depth EDA on 750k rows / 16 features, identifying severe class imbalance (~7.3:1) and disguised missing values ("unknown" categories, pdays = -1).
• Diagnosed and removed a documented temporal data leakage on the duration feature (point-biserial r = +0.52), favouring production realism over leaderboard score.
• Built a leakage-safe Scikit-learn Pipeline (ColumnTransformer + StandardScaler + OneHotEncoder) fitted only on training folds.
• Benchmarked six models (Logistic Regression, LDA, QDA, Gaussian NB, Random Forest, XGBoost) on four complementary metrics (ROC-AUC, PR-AUC, F1, Brier), validated by stratified 5-fold cross-validation.

Code Presentation

IMDB Sentiment Classification

Binary text classification using NLP pipelines and various models (MLP, LSTM, TextCNN, LR, SVM) to predict movie review sentiment.

Python, PyTorch, Scikit-learn, NLP

View Details

School: UTBM

Team size: 3

Date: 2026

Course: DS52 Machine Learning

A comparative study of traditional ML and deep learning models for text sentiment analysis on the 50K IMDB dataset:

• Implemented data cleaning (regex, HTML/URL removal) and vectorization pipelines using TF-IDF (with custom stoplist to preserve negations) and integer sequence padding.
• Benchmarked traditional models (Logistic Regression, SVM) against neural architectures (3-layer MLP, Unidirectional LSTM, and Kim-style TextCNN).
• Demonstrated that a simple TF-IDF representation combined with an MLP (90.5% accuracy) or LR/SVM out-performed a from-scratch LSTM (84.9%) which suffered from limited embedding representation.
• Analyzed feature interpretability by identifying the most discriminative bigrams and evaluated training speed versus accuracy.

Code PDF Report

Teeko - AI Strategy Game

A Python implementation of the Teeko board game featuring a responsive GUI and a scalable AI opponent using advanced search algorithms.

Python, Tkinter, Minimax, Alpha-Beta Pruning

View Details

School: UTBM

Team size: 3

Date: 2025

Course: Artificial Intelligence

This project involved developing a full version of the Teeko strategy board game with a strong focus on Artificial Intelligence:

• Developed a graphical user interface (GUI) using Python and Tkinter to manage game phases (Placement and Movement).
• Implemented a scalable AI with three difficulty levels, culminating in a "Hard" mode powered by the Minimax algorithm with Alpha-Beta pruning.
• Optimized decision-making (Depth 4) using heuristic evaluation functions to assess board control, alignment threats, and win conditions.
• Solved complex state management challenges to handle move validation and real-time game updates efficiently.

Report (french) Code

Calorie Crew - Social Nutrition Tracker

A native Android application designed to assist users in managing their health through personalized calorie tracking and community-driven challenges.

Kotlin, Jetpack Compose, Firebase, MVVM

View Details

School: UTBM

Team size: 4

Date: 2024

Course: Mobile Development

Calorie Crew is a mobile application developed to combine fitness tracking with social motivation[cite: 18]:

• Built entirely with Kotlin and Jetpack Compose, following Material 3 design principles for a modern and reactive UI.
• Structured using the MVVM architecture (Model-View-ViewModel) to ensure scalability and clean separation of concerns.
• Leveraged Firebase for secure authentication and Cloud Firestore for real-time data synchronization of meals, profiles, and messages.
• Implemented advanced features such as dynamic calorie calculation, community creation, real-time chat, and gamified challenges.

Report Figma

Moodneuille - a simplified educational platform

A comprehensive web application inspired by Moodle, featuring user role management, course administration, and dynamic content interaction.

Symfony 7, MySQL, Bootstrap, AJAX

View Details

School: UTBM

Team size: 2

Date: 2024

Course: Web Development (WE4A)

In this project, we built a full-stack educational platform using the Symfony 7 framework:

• Designed the database using the Merise methodology (MCD/MLD) and implemented it via Doctrine ORM.
• Structured the application using the MVC architecture to manage complex user roles (Administrator, Professor, Student).
• Implemented secure authentication and access control to manage permissions for courses and content.
• Integrated JavaScript and AJAX to enable dynamic interactions, such as deleting posts without page reloads.

Report (french) Code

UTBMachine - Logic Game Adaptation

A digital adaptation of the "Turing Machine" board game featuring a custom UI and a robust object-oriented architecture.

Java, Swing, MVC, UML

View Details

School: UTBM

Team size: 2

Date: 2024

Course: Object-Oriented Design (AP4B)

UTBMachine is a desktop application developed to re-imagine the board game "Turing Machine" with a custom university theme:

• Developed a complete graphical interface using Java and the Swing library to simulate the game board and note sheets.
• Implemented the MVC pattern (Model-View-Controller) to strictly separate the game logic from the UI, ensuring modularity.
• Designed the system using advanced UML modeling (Use Case, Class, Sequence, and State-Transition diagrams).
• Created 6 distinct logic scenarios with varying difficulty levels, allowing for solo, competitive, or cooperative gameplay.

Report (french) Code

The majority element algorithm

Study about the majority element problem and implementation of a naive quadratic method and a more efficient divide-and-conquer version.

Jupiter, Pyhon

View Details

School UTBM

Team size: 3

Date: 2024

Course: Classical and Quantum Algorithm

In this project :

• We studied the majority element problem and implemented two algorithms to solve it: a naive quadratic method and a more efficient divide-and-conquer version.
• We proved the correctness of the divide-and-conquer algorithm (termination and partial correctness) and compared the time complexities of both approaches.
• Finally, we performed experiments showing that the divide-and-conquer algorithm significantly outperforms the naive one on medium and large datasets.

Report Video

The No Communication Theorem

Study about Quantum Entanglement. Can we communicate faster that the speed of light using quantum computing ?

Jupiter, Python, MPQP

View Details

School: UTBM

Team size: 3

Date: 2025

Course: Classical and Quantum Algorithm

In this project, we demonstrate both mathematically and experimentally, that quantum entanglement cannot be used to transmit information:

• Prepared a maximally entangled Bell state using the MPQP quantum circuit simulator.
• Measured Alice’s qubit in different bases (Z and X) to test for signaling effects.
• Measured Bob’s qubit independently and analyzed the statistical outcomes.
• Proved mathematically using reduced density matrices that Bob’s local state is unaffected by Alice’s choices.
• Confirmed through simulation that Bob’s measurement results remain 50/50, regardless of Alice’s measurement basis.

Video

Certifications

Getting started with Deep Learning

NVIDIA

Fundamental concepts of Deep Learning, covering neural network architectures and implementation strategies.

View Certification

Building LLM Applications With Prompt Engineering

NVIDIA

Advanced techniques for leveraging Large Language Models through effective prompt engineering.

View Certification

Data Visualization with Python

IBM

Mastering data visualization libraries like Matplotlib, Seaborn, and Folium to derive insights from data.

View Certification

Google Cloud Data Analytics Certificate

Google

Mastering foundational and advanced data analytics skills, including data cleaning, analysis with SQL and BigQuery, and data visualization to drive business decisions.

View Certification

Certificate of practice of professional networking

MyJobGlasses

4 meetings with professionals conducted on My Job Glasses from November 1, 2022 to June 12, 2023

View Certification

Business Linguaskill

Cambridge Assessment English

Demonstrated fluency in business English, covering reading, listening, writing, and speaking skills.

Score : 175/180

View Certification

French High School Degree - Mathematics

Baccalauréat

Score : 19/20

View Certification

French High School Degree - Physic and Chemistry

Baccalauréat

Score : 18/20

View Certification