A controlled study of compact neural reranking for Sonja PhraseFlow. Two training objectives and three seeds are compared on the same 600 reused development cases, with learning curves, paired uncertainty intervals and native R/CPU verification. The findings support retaining the existing model. This is exploratory research, not an independent final-test accuracy claim. Author: Sonja Sahebzad, Utrecht, the Netherlands.

A five-slide RStudio Presenter pitch for Sonja PhraseFlow 1.4: a public Shiny app with twelve selectable languages, 5/10/20 word choices, an illustrated user guide and measured accuracy with uncertainty intervals. Additional language models and short continuations are experimental. Author: Sonja Sahebzad, Utrecht, the Netherlands.

A reproducible case study by Sonja Sahebzad, separating a 100% reviewed outcome from original model performance. Includes aggregate results, interactive charts, discussion, conclusions and independent evidence on confidence and answer coverage. Assessment questions and answers are excluded.

Five-slide RStudio Presenter pitch for the Johns Hopkins University / Coursera Developing Data Products course project. It presents the Urban Air Quality Explorer Shiny application, its interactive features, model comparison, and limitations. Created by Sonja Sahebzad on September 17, 2026.

An interactive R Markdown presentation exploring urban air quality with Plotly. It includes interactive charts, interpretations, limitations, visible slide navigation, and the creation date.

An interactive Leaflet map created in R Markdown featuring selected learning, cultural, and green locations in Amsterdam. The map includes clickable markers, informative pop-ups, a legend, thematic layers, and API-free OpenStreetMap tiles.

An interactive city guide created by Sonja Sahebzad using R Markdown and Leaflet. The map highlights selected learning, science, cultural, and green locations in Amsterdam with colour-coded markers, informative pop-ups, a legend, and selectable map layers.

Practical Machine Learning course project by Sonja Sahebzad. A random forest model was developed using five-fold cross-validation to predict barbell-lift execution from wearable-sensor data. The report includes preprocessing, model evaluation, expected out-of-sample error, discussion, conclusions, and predictions for 20 test cases.

A regression analysis of the mtcars data examining whether manual or automatic transmission is associated with better fuel economy and quantifying the adjusted MPG difference.

Analysis of NOAA storm data for Course Project 2 in Reproducible Research. The report compares fatalities, injuries, and property and crop damage, and includes R code, two figures, limitations, and a conclusion.

The assignment involves addressing 5 key questions related to processing and analyzing “activity monitoring data.” This data analysis project revolves around the exploration of activity monitoring data collected over two months. The dataset originates from a study in which an individual used a personal activity monitoring device to track daily step counts at five-minute intervals. The dataset includes variables such as steps taken, date, and time intervals.