Pascal Nague
Computer Science student specializing in Machine Learning and Data. I develop practical AI solutions, analyze data, evaluate models and optimize machine learning pipelines with Python and PyTorch.
I’m Pascal Nague, a Computer Science student at RPTU Kaiserslautern with a strong focus on Machine Learning, Data and AI systems.
I enjoy understanding how intelligent systems work — from the data they consume to the way their performance, reliability and efficiency are measured.
My work combines Python, PyTorch, data analysis, machine learning experimentation and performance benchmarking. My completed Bachelor Project at DFKI focused on efficient PyTorch data loading, concurrent pipelines and reproducible performance measurement, followed by external validation on Food-101.
I am especially interested in building practical systems that are measurable, understandable and useful in real-world environments.
I work on practical problems across data, machine learning and AI evaluation. My focus is on building measurable solutions, understanding model behaviour, automating data workflows and improving the performance of ML pipelines.
Cleaning, exploring and structuring datasets so that raw data becomes reliable, understandable and ready for analysis or machine learning.
Building initial classification, regression or clustering models, establishing baselines and evaluating whether an ML approach is useful.
Measuring model performance using appropriate metrics, error analysis and structured comparisons to understand where a model succeeds or fails.
Designing test cases, evaluation criteria and error categories to assess chatbot and LLM behaviour across realistic prompts and edge cases.
Automating repetitive data tasks such as merging files, transforming exports, calculating metrics and generating repeatable reports.
Benchmarking data-loading pipelines, comparing DataLoader configurations and identifying bottlenecks that affect throughput and model training.
My technical background combines software development, data analysis, machine learning and performance-oriented ML systems.
Python · Java · SQL · HTML · CSS · JavaScript
Data Cleaning · Exploratory Data Analysis · Data Visualization · Numerical Computing
Classification · Regression · Model Evaluation · Cross-Validation · Error Analysis
PyTorch · CNNs · YOLOv8 · OpenCV · Image Classification · Training · Validation · Image Preprocessing
PyTorch DataLoader · Data Pipelines · Concurrency · Multithreading · Multiprocessing · Throughput Benchmarking · Bottleneck Analysis
Test Case Design · WER/CER · ASR Evaluation · Error Classification · Quality Assessment
Selected engineering projects across ML systems, AI evaluation, computer vision, end-to-end ML applications and performance benchmarking.
A lightweight experimental library for Parquet-backed image datasets with a custom multi-threaded DataLoader, throughput instrumentation and reproducible benchmarks. The optimized loader reached a measured peak of 3,245 samples/s on the project benchmark and was published as a PyPI package.
An engineering and learning project for evaluating French speech-to-text outputs and reviewing language-model responses. The current foundation includes manifest validation, ASR pilot tooling, WER/CER metrics, reporting and automated tests.
External validation of the custom PyTorch data-loading library on a reproducible 10,000-image Food-101 workload. The experiments compare scaling across worker counts, resolutions and batch sizes and demonstrate that DataLoader performance is workload-dependent.
A modular real-time computer-vision application built with YOLOv8 and OpenCV, including webcam streaming, person filtering, configurable alert logic, automated screenshots and runtime statistics.
An end-to-end machine-learning application covering data generation and feature engineering, Random Forest modelling, MLflow experiment tracking, a FastAPI prediction API, a Streamlit interface and Docker-based packaging.