Computer Science student at RPTU Kaiserslautern-Landau, working on machine learning systems, data pipelines, model evaluation and performance engineering.
I am particularly interested in the engineering around ML models: how data is loaded and validated, how model behaviour is measured, where performance bottlenecks appear, and how experimental ML components become usable software.
Portfolio: pascal-cabrel-nague-portfolio.netlify.app
LinkedIn: linkedin.com/in/pascal-nague
- ML systems and performance engineering
- Data pipelines and data quality
- ASR / LLM evaluation
- Computer vision applications
- AI-assisted software and automation
- Benchmarking, profiling and reproducible experiments
I prefer projects where results can be measured, reproduced and explained, rather than treating the model as a black box.
A lightweight experimental data-loading library developed during my completed Bachelor Project at DFKI.
The project explores Parquet-backed image datasets and a custom concurrent loading architecture designed to integrate with standard PyTorch training code.
I built and benchmarked:
- a Parquet-backed dataset implementation
- a multi-threaded loading pipeline
- worker and batching orchestration
- throughput instrumentation
- scaling experiments from 1 to 32 workers
- comparisons against
torch.utils.data.DataLoader
The original benchmarks reached a measured peak of approximately 3,245 samples/s at 16 workers.
To test whether those results generalized beyond the original workload, I later built a separate validation benchmark using Food-101. That experiment showed an important limitation: the custom loader remained competitive at low worker counts, while PyTorch scaled better under the new workload.
Stack: Python · PyTorch · PyArrow · Parquet · Concurrency · Benchmarking · SLURM
Repository · Food-101 validation
An active engineering project for evaluating speech-to-text systems and reviewing LLM outputs.
The project separates three different quality problems:
- Data quality — is the reference itself correct?
- Model quality — what errors does the model produce?
- Software quality — does the evaluation system measure and store results correctly?
The current implementation includes:
- Python package and CLI
- JSON schema validation
- audio manifest validation
- ASR pilot tooling
- WER / CER computation
- automated tests
- reporting utilities
- recording and reference-review workflows
- local model and hardware preparation tools
The revised peer beta is currently under development.
Stack: Python · ASR · WER/CER · pytest · JSON Schema · Data Validation · Model Evaluation
A modular real-time computer-vision application built with YOLOv8 and OpenCV.
Instead of stopping at model inference, the project wraps the detector inside an application with separate components for:
- webcam streaming
- YOLO inference
- person filtering
- configurable alarm logic
- automated screenshots
- FPS and runtime statistics
It was mainly an exercise in moving from using a pretrained model to integrating one into a structured software system.
Stack: Python · YOLOv8 · OpenCV · Computer Vision
An external validation experiment for the Minimal Dataset project using a reproducible 10,000-image subset of Food-101.
The benchmark compares the custom loader with PyTorch across:
- multiple worker counts
- different image resolutions
- different batch sizes
- identical preprocessing pipelines
The results helped identify where the custom architecture scales well and where its current design reaches a throughput plateau.
Stack: Python · PyTorch · Food-101 · Parquet · Benchmarking · Performance Analysis
Machine Learning
Python · PyTorch · Scikit-learn · YOLO · OpenCV
Data & ML Systems
Pandas · PyArrow · Parquet · Data Pipelines · Model Evaluation · Benchmarking · Concurrency
Software Engineering
Git · GitHub · Docker · Linux · Windows · pytest · SLURM
Web
HTML · CSS · JavaScript
My portfolio is built from scratch with vanilla HTML, CSS and JavaScript.
It serves as a central place for my technical projects, current engineering focus and professional profile.
- B.Sc. Computer Science — RPTU Kaiserslautern-Landau
- Bachelor Project at DFKI — completed
- Speech & LLM Quality Lab — active development
- Building deeper experience in ML systems, evaluation, data engineering and AI automation
- Open to Werkstudent roles and technical projects in ML, Data and AI Engineering
Build → measure → validate → improve.