aleth was born out of a research innitiative to lower the barrier to synthetic sensor telemetry in buildings.

Usage

aleth runs on top of ollama for local inference. Start the server with an extended context window before running anything:

OLLAMA_CONTEXT_LENGTH=64000 ollama serve &
ollama pull gpt-oss:20b

Install the package from src/aleth/, then point it at a scenario in plain English:

cd src/aleth && pip install -e .

aleth --scenario "CO2 sensor in a university lecture hall, heavy occupancy on weekdays" \
      --start-year 2024 --years 2 --freq-minutes 30

Each run writes a timestamped folder under results/ with a CSV of the generated timeseries, a JSON of the inferred value ranges, and a set of diagnostic plots. The model and ollama endpoint can be changed in config.py.

# a few more examples of what the scenario argument can express
aleth --scenario "Temperature sensor on a rooftop HVAC unit in Madrid"
aleth --scenario "Water conductivity sensor in a building's cooling tower"
aleth --scenario "PM10 air quality sensor near a busy urban road, rush-hour spikes"

Citation

To cite this work, feel free to use the following BibTeX entry:

@inproceedings{petrescu2026aleth, 
  title={Generative Models as a Catalyst for Lowering the Barrier to Synthetic Sensor Telemetry},
  author={Petrescu, Stefan and Rellermeyer, Jan S.},
  year={2026},
  booktitle = {Proceedings of the 13th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation},
  series = {BuildSys '26}
}