M.Sc. Data Science, TU Wien · Vienna
Kasra
Zarinehbaf Asadi
Data Scientist
Research assistant at WU Wien from 2024 to 2025, building machine learning, forecasting and NLP models for international business research, and thesis research at the Complexity Science Hub reconstructing capital flows through a year of financial transaction data.
Projects
RAG for Austrian law
Retrieval-augmented question answering over 17 Austrian federal statutes, in German. Every change is measured against a frozen test set of 776 questions taken from court decisions, where both the question and the governing paragraph were written by lawyers. One result from that work: a generated evaluation set overstated improvements by 4 to 17 times compared with the human-labelled one.
- 0.439
- recall@5
- 0.323
- full_recall@5
- 22 ms
- latency
Try it — soonCode
Austrian electricity price forecast
A forecasting pipeline built to production rules: inputs frozen at the 10:30 Vienna cutoff so nothing from the delivery day leaks in, versioned snapshots of every input, and scoring that keeps each price revision as its own artefact. The numbers below come from an eight-day baseline run that proves the pipeline works end to end — not from a model with demonstrated skill.
- 34.84
- MAE €/MWh, smoke test
- 48.72
- RMSE €/MWh, smoke test
- 768
- intervals scored
Weekly baseline with scheduled daily forecasts and historical replay results.
App for todos, notes and workouts
One calendar board with three layers in the same day columns: todos that follow you into today until they're done, written notes that stay on the day they were written, and workouts with progress charts across the weeks. Click the wordmark, top left, to switch between them.
- 3
- layers
- 113
- tests
- 0
- servers
The demo opens filled with example data, including ten weeks of training. Change anything you like — nothing is saved, and a refresh puts it back.
