01
Decision platforms
Reusable foundations connecting trusted data, modelling workflows, evaluation, and decision-facing outputs.
Platform building
Data Scientist · Luxembourg
Kaan Kaya / Profile 2026
Models for the real world.
I build data products that move from prediction to explanation to causality—so operational teams can make better, defensible decisions.
DECISION SYSTEM
From signals to evidence
observedbenchmark
01
Predict
LGBM02
Explain
SHAP03
Estimate
DML10+years across data, finance & risk
4linked layers: data → decision
1principle: useful models earn trust
01 / PRACTICE
Selected capabilities
Building systems for evidence-led decisions.
Across my work and independent practice, I focus on reliable platforms, predictive modelling, explainability, and causal reasoning.
02
Decision modelling
Predictive workflows designed to support scenario analysis and informed operational choices.
Predictive analytics
03
Explainable ML
Transparent diagnostics that help stakeholders understand important patterns and model limitations.
Interpretability
04
Causal evidence
Structured methods for reasoning about effects, assumptions, uncertainty, and robustness.
Causal inference
02 / APPROACH
A model is only one layer.
The useful unit is a decision system: reliable inputs, strong prediction, honest explanations, causal assumptions, and an interface people can act on.
“Make uncertainty visible. Make reasoning inspectable. Make the next action clearer.”
Prediction
Learn the operational pattern
Attribution
Expose the model's reasoning
Causality
Test what may change the outcome
Decision
Ship evidence into a usable system
03 / SELECTED WORK
Built to learn.
Built to run.
01020304
DeFtunes Data Pipeline
An end-to-end analytical platform built across orchestration, transformation, warehousing, and BI.
AWSAirflowdbtRedshift
AI Code Arena Quest
An interactive learning product that turns machine-learning concepts into hands-on practice.
ReactMachine LearningAI EducationFull-Stack
PyTorch for Deep Learning
A practical body of work spanning modern neural architectures, optimization, and deployment.
PyTorchDeep LearningTorchVisionOptuna
LectureNotes RAG System
A GPU-accelerated retrieval system for grounded question answering over technical material.
RAGFAISSGPU AccelerationPyTorch
04 / JOURNEY
Context is a feature.
Experience across big tech, fintech, asset management, and banking shapes how I frame risk, cost, and operational decisions.
Jan 2025 - Present
Data Scientist
AmazonBuilding data products and analytical tools for operational decision support.
Data ScienceMachine LearningExplainable AICausal InferencePlatform Engineering
Aug 2024 - Jan 2025
Senior Data Analyst (Head of Data)
HubfinanceDesigned and automated ETL/ELT pipelines to consolidate product and customer data into a central analytics store
May 2022 - Jul 2024
Technical Business Analyst (Data Analyst)
SchrodersLed the redesign of financial and regulatory reporting processes, migrating manual reports to automated data pipelines
Aug 2021 - Apr 2022
Data Consultancy & Contract Work
Lingaro, CarmignacSupported clients with data management and automation solutions
May 2020 - Jul 2021
Technical Business Analyst (Associate Director)
Standard Chartered BankLed initiatives to redesign market risk reporting systems, consolidating data from multiple trading and risk platforms
05 / CONTACT
Let’s make the question sharper.
I’m always interested in rigorous data science, causal inference, and the craft of turning complex models into useful products.
kaankaya@live.com