Open to Research Collaborations

Ossama Sabti

Passionate about machine learning, reinforcement learning, and building intelligent systems. I combine mathematical rigor with hands-on engineering to solve real-world problems.

oussamasabti6@gmail.com Rabat, Morocco · Open to relocation github.com/ossamasabti
Ossama Sabti
Profile

About Me

Background

Holder of a Master's degree in Mathematics, Data Analysis, and Learning (Data Science track) from the Faculty of Sciences, Rabat (UM5). I bring a strong background in applied mathematics, optimization, and statistical learning to the field of AI. Currently a PhD candidate at CEDoc INSEA, Rabat.

Research Direction

Highly motivated to develop machine learning and reinforcement learning methods for modeling, prediction, and control of real-world dynamical systems.

Tech Stack

Tools & Expertise

Data Science

PythonPandasNumPyMatplotlibSeabornScikit-learn

Deep Learning

TensorFlowKerasPyTorchVision Transformer

Dev & Tools

DjangoSQLPostgreSQLGitLaTeX

ML & BIM

XGBoostONNXRevitAutoCAD
Background

Experience & Education

Work

2026
CAD to BIM Intern
BIMPioneers, Casablanca
  • Final-year project (BIMFlux): automated conversion of 2D CAD plans into 3D BIM models.
  • ML detection and classification of drawing elements.
2026
Organizer, MathICs-Club
Faculty of Sciences, Rabat (UM5)
  • Led logistics for International Day of Mathematics event (200+ attendees).

Education

2026 – present
PhD Candidate
CEDoc INSEA, Rabat
2024 – 2026
Master's in Mathematics, Data Analysis & Learning
UM5 Rabat
2024
Bachelor's in Applied Mathematics
UCD El Jadida
2023
CPGE (MP) Mathematics & Physics
Lycée Mohammed V, Casablanca · Admissible to CNC 2023
2020
Baccalaureate Mathematical Sciences B
Lycée Bismilliah, Settat
Portfolio

Featured Projects

2026
BIMFlux: HVAC Ducts from AutoCAD 2D to Revit
XGBoost · ONNX · Python · Revit
  • Detection and automatic generation of HVAC ducts from 2D AutoCAD plans into a Revit BIM model
  • Edge classification on a candidate graph with XGBoost, combined with a geometric score and a rule engine
  • Python MCP server connecting Revit to Claude for model diagnostics
2025
AI vs Human Art Classification
Vision Transformer · PyTorch
  • 84% accuracy distinguishing human art from AI
  • Fine-tuned pretrained transformer for domain adaptation
View on GitHub
2024
Employability of Young Graduates
Scikit-learn · Pandas
  • Predictive modeling & employment barrier analysis
  • Survey data cleaning and classification to predict hiring likelihood
View on GitHub
2024
E‑recruitment Platform
Django · PostgreSQL
  • Full-stack job portal with resume upload and tracking
  • Authentication system and admin dashboard
View on GitHub
Academic

Research Presentations

2025
Seminar Presentation
Topological Autoencoders++ : Fast and Accurate Cycle-Aware Dimensionality Reduction
Persistent Homology · Topological Data Analysis · Autoencoders · PyTorch

Presented a critical analysis of the paper by Clémot, Digne & Tierny (arXiv:2502.20215) in the Master MADA seminar at UM5 Rabat. The work extends Topological Autoencoders (TopoAE) to faithfully preserve 1-dimensional topological cycles.

Persistent HomologyDimensionality ReductionAutoencoderTopological Data AnalysisarXiv:2502.20215
Languages

Languages

Arabic (Native)French (C1)English (B2)
Connect

Get In Touch

Let's Collaborate

Open to research collaborations, academic exchanges, and opportunities in data science and AI.