Vincent SCHULER/Data Scientist

Portfolio

Welcome to my portfolio.

Explore around twenty of my Data Science projects.

Confidentiality: All data has been anonymized and identifying details have been altered.

01

Predictive Machine Learning

Predictive Machine Learning estimates the likelihood of future customer behavior, such as a subscription, cancellation or vehicle purchase, so teams can target the right actions.

My approach

  • Business framing, feature engineering and time-aware validation aligned with the real cost of prediction errors.
  • Explainable scores that integrate directly into campaigns, decision processes and customer journeys.
  • Scikit-Learn
  • LightGBM
  • XGBoost
  • CatBoost
  • Vertica
  • Pandas
  • Polars

02

Time Series

Time-series analysis can forecast future volumes or detect precisely timed events in a signal, from call peaks to sleep phases.

My approach

  • Rigorous time-based validation and features that capture seasonality, events and changing patterns.
  • Forecasting and detection models evaluated with metrics that reflect the practical consequences of errors.
  • Scikit-Learn
  • LightGBM
  • XGBoost
  • CatBoost
  • Pandas
  • Polars

03

Fraud Detection

Effective fraud detection isolates the few genuinely suspicious cases without overwhelming investigation teams with false alerts.

My approach

  • A pragmatic combination of business rules, supervised learning and anomaly detection.
  • Thresholds and metrics calibrated to financial exposure and investigation capacity.
  • Scikit-Learn
  • LightGBM
  • Clustering
  • Isolation Forest
  • Autoencoders

04

Recommendation Systems

Search and recommendation systems help each user find the most relevant products, content or automation rules.

My approach

  • Ranking, similarity and hybrid methods adapted to the available data and use case.
  • Evaluation focused on relevance, diversity, conversion and business constraints.
  • Scikit-Learn
  • LightGBM
  • XGBoost
  • CatBoost
  • Pandas
  • Polars

05

Entity Resolution & Data Quality

Entity resolution reconciles names, addresses and records despite spelling errors, duplicates and incompatible formats.

My approach

  • Candidate-generation and similarity methods designed to scale to millions of records.
  • Normalized, deduplicated and enriched data with controllable confidence levels.

06

NLP & Text Analytics

Natural language processing turns unstructured text into useful information: classifications, entities, scores and actionable signals.

My approach

  • Lightweight methods or transformer models, selected according to the complexity the problem actually requires.
  • Large-scale preprocessing, robust validation and integration into production workflows.
  • TF-IDF
  • Transformers
  • DeBERTa
  • NLTK
  • TensorFlow
  • PyTorch

07

Computer Vision

Computer vision extracts useful information from images and video to detect, classify, measure or anonymize content, including in real time.

My approach

  • End-to-end pipelines, from data preparation and augmentation to model evaluation.
  • Architectures and inference times adapted to image quality and operational constraints.
  • OpenCV
  • TensorFlow
  • PyTorch
  • YOLO
  • EfficientNet
  • Albumentations

08

Optimization

Optimization identifies the best achievable trade-off when resources are limited and several objectives compete.

My approach

  • Explicit modelling of objectives, constraints and the business cost of each trade-off.
  • Custom heuristics for routing, scheduling and resource allocation at scale.
  • LKH-3
  • Clustering
  • Pathfinding
  • Heuristic optimization
Interactive visualization
Interactive case study

Optimizing City Allocation among Transporters

I allocated 350 German cities among 15 transporters to reduce total travel while keeping routes and workloads balanced.

Video case study

The Christmas Card Conundrum

Working with four teammates, I optimized the path of a robotic arm across every pixel of an image. Our solution reached 99.998% of the estimated optimum and placed 9th out of 874 teams.

Optimizing Search Queries
Case study

Optimizing Search Queries

An optimization pipeline generates concise search queries that retrieve precise sets of target patents from a corpus of 13 million documents.