01 / PROJECT
iRacing podium prediction
Predicting the iRacing podium as a race unfolds. A Python pipeline combines live telemetry with driver history, comparing three machine learning models to find the best fit for live use.
TECHNICAL CONSULTANT & DEVELOPER · MALTA
Building ideas
into useful
software.
I’m Antoine, a software developer and ERP technical consultant based in Malta. Here you’ll find my projects in software and machine learning, starting with a live iRacing podium predictor.
View CV PDFSoftware developmentData & machine learningERP development
FEATURED PROJECT / 0012025
Race data.
Podium intelligence.
Podium intelligence.
Live telemetryiRacing SDK
Driver historyiRacing Data API
Hybrid prediction pipelinePython · SQL Server · ML
3sec
XGBoostRankNetRandom Forest
02
01PODIUM
03
PYTHON/JAVASCRIPT/SQL/PYTORCH/SCIKIT-LEARN/GIT
01 / PROJECTS
Software, experiments & research.Ideas turned
into working projects.
What I built, how it works, and what I learned along the way.
PROJECT NOTES · BSc (HONS) DISSERTATION
From telemetry
to the podium.
The goal was to find a model that could keep up with a changing race. I built the data pipeline, compared the models and connected the chosen RankNet model to a live prediction loop.
Predictions between lap updates.
Combine live telemetry with driver history to predict podium finishes as race conditions change.
A complete prediction pipeline.
Connected the iRacing SDK and Data API to SQL Server and Python, with three-second updates and SHAP explanations.
Accuracy meets adaptability.
Compared three models on 18 unseen races. XGBoost led test accuracy; RankNet was chosen for live mid-lap adaptability.
Model results & development walkthroughs
MODEL COMPARISON
Three models.
Different strengths.
Reported top-3 test accuracy from the dissertation.
PAIRWISE NEURAL RANKING
Adaptable between lap updates.
Chosen for mid-lap adaptability and a small train–test gap. Separate live mid-lap testing averaged approximately 75% top-3 accuracy.
GRADIENT-BOOSTED TREES
Strong at race snapshots.
Highest reported test top-3 accuracy, with higher train–test variance and less flexibility between lap updates.
BAGGED DECISION TREES
A stable baseline.
Lowest reported mean absolute error, with weaker podium ranking and less suitable inference behaviour for the live loop.
Test figures and separate live results measure different conditions. Limited weather diversity and missing tyre-wear and brake-temperature data constrain generalisation. See section 4.2.2 of the full thesis.
02 / EXPERIENCE
More about meWhere I’ve put it into practice.
JUNIOR TECHNICAL CONSULTANT · CURRENT
KPMG
Microsoft Business Solutions Malta
ERP development in Microsoft Dynamics 365 Finance & Operations, writing X++ to support business systems.
SOFTWARE ENGINEER · APPRENTICESHIP
PTL Ltd
Software development with C#, SQL and ASP.NET, alongside work with Microsoft Dynamics Business Central.
SYSTEM SOFTWARE DEVELOPER · CONTRACT
MECB Ltd
Built and maintained internal tools, working with JavaScript, web hosting and technical troubleshooting.
For the full career history, view my CV PDF.
03 / CERTIFICATES
Professional qualifications.
Dynamics 365 Fundamentals (ERP)
Microsoft Certified · November 2025
Certified Associate: Game Developer
February 2024 · Expires February 2027
04 / GET IN TOUCH
MALTA · CONNECTED EVERYWHEREHave something
in mind? Let’s talk.
For technical conversations, software projects, or new collaborations.