Language-model evaluation
Bachelor’s thesis work compared quantized and full-precision models and examined task-level results using statistical analysis; detailed source and results are confidential.
What I’ve used in professional work, independent builds, guided projects and academic work. Each skill links to the evidence behind it.
35 skills shown
Ingest, validate and transform data with Python and SQL. Build pipelines and relational datasets that others can reuse.
Used for data preparation, project pipelines, and workplace automation.
Evidence
Applied to relational data transformations and reporting models.
Evidence
Used for layered warehouse transformations and time-series data storage.
Evidence
Used to prepare procurement, logistics, and project data.
Evidence
Prepared SQL checks for warehouse data and reporting views; studied data quality in MSc coursework.
Evidence
Defined telemetry topics and a producer for a simulation.
Evidence
Wrote Structured Streaming transformations for simulated telemetry.
Evidence
Configured Parquet sinks and checkpoint locations for a streaming prototype.
Evidence
Acquired, combined and preprocessed data with Python and Pandas in coursework.
Evidence
Studied relational and NoSQL database systems.
Evidence
Replace repetitive steps with Python scripts, document-processing workflows and internal tools. Use Power Platform where it fits.
Built automations for recurring operational workflows.
Evidence
Built internal tools to support operational work.
Evidence
Automated data processing for operational and sustainability reporting.
Evidence
Developed Python workflows for information extraction, validation, and structured output.
Evidence
Built an interactive tool for probing language models in my bachelor’s thesis.
Evidence
Train and evaluate forecasting models, build retrieval applications, and test where those systems fall short.
Built 48-hour energy forecasts with LightGBM.
Evidence
Integrated experiment tracking in the energy forecasting project.
Evidence
Compared language-model variants and studied model selection and validation.
Evidence
Used transformer models in thesis and team project work.
Evidence
Built a prototype that combines keyword and vector search, reranks results, and prompts answers to cite the rulings they draw from.
Evidence
Used in an HSLU team project for student-summary scoring.
Evidence
Used in the neural-model paths of an HSLU summary-scoring team project.
Evidence
Coursework covered deep learning and neural networks.
Evidence
Coursework covered text processing, sentiment analysis, document classification, and deep-learning NLP.
Evidence
Studied reliable, reproducible machine-learning systems.
Evidence
Model data for reporting, check that the numbers reconcile, and make results usable in dashboards.
Structured warehouse reporting views as a star schema.
Evidence
Maintain and improve dashboards and reports for operational quality and safety.
Evidence
Used in procurement and logistics data preparation.
Evidence
Applied statistical analysis in thesis research and studied classical and Bayesian methods.
Evidence
Covered in Programming for Data Science coursework.
Evidence
Used R and RStudio for statistical analysis in coursework.
Evidence
I’m interested in how components fit together and how design choices affect reliability and maintainability.
Connected ingestion services, storage and forecast outputs; documented a FastAPI service for local use.
Evidence
Documented how raw data, transformations and reporting views fit together.
Evidence
Version control and reproducible development environments support my project work.
Used in project environments and covered in MSc software-engineering coursework.
Evidence
Version control across public projects and MSc software-engineering coursework.
Evidence
Academic evidence
Completed coursework and selected thesis and team-project evidence.
2025 – present
In progress, with coursework in data integration, data quality, and predictive modelling.
I’m researching how to make product carbon footprint calculations easier to update and trace with a modular, semi-automated approach.
2021 – 2025
Foundation in machine learning, NLP, statistics, databases, MLOps, and software engineering. My thesis combined reproducible evaluation of quantized and full-precision language models with a usable probing interface.
Thesis: “Automated Assessment of Linguistic Abilities of Language Models”.
Bachelor’s thesis · 6.0/6.0
Bachelor’s thesis work compared quantized and full-precision models and examined task-level results using statistical analysis; detailed source and results are confidential.
Bachelor’s thesis work compared quantized and full-precision models and examined task-level results using statistical analysis; detailed source and results are confidential.
I’m researching how to make product carbon footprint calculations easier to update and trace with a modular, semi-automated approach.
Repository documents an HSLU team project using a three-path ensemble; individual implementation contributions are not specified.
Data acquisition, integration, and preprocessing with Python and Pandas, alongside data-quality assessment and improvement.
Database coursework covered relational design, SQL, and NoSQL.
Coursework covered predictive modelling, model validation, and statistical methods.
Coursework covered deep learning, NLP, and reproducible machine-learning systems.
HSLU AI Challenge group assignment documented by the team repository; README describes a three-person team and a three-path ensemble combining ROUGE/neural-model features, LightGBM, and DeBERTa v3 Large.
Data acquisition, integration, and preprocessing with Python and Pandas.
Data-quality dimensions, assessment, and improvement.
Coursework covering Git, Docker, and software engineering in Python.
Regression, classification, resampling, model selection, and regularization.
Dimensionality reduction, clustering, classification, and ensemble methods.
Probability, hypothesis tests, regression, Bayesian models, and exploratory analysis using R and RStudio.
Python, NumPy, SciPy, Matplotlib, Pandas, and machine-learning fundamentals.
Database application design, schemas, data transformation, SQL, and NoSQL.
Supervised and unsupervised learning, deep learning, generative models, and transfer learning.
Text processing, semantic and sentiment analysis, document classification, and deep-learning NLP.
Reliable, scalable, reproducible machine-learning systems and deployment.
Next steps
A few areas I’d like to explore next.
Planned
SQL transformations and data testing.
Planned
Lakehouse workflows and data processing.
Planned
Cloud data services.