Tarlan Bahadori

Tarlan Bahadori

Computer Science PhD Student

LASEK: LLM-Assisted Style Exploration Kit for Geospatial Data

Published in VLDB 2025 — Demo Track (London), 2025

Authors: Tarlan Bahadori, Sai Sreekar Sarvepalli, Ahmed Eldawy

LASEK is a system for interactive exploration and styling of large geospatial datasets.
Users can upload their own data, get LLM-generated style suggestions (e.g., graduated ramps, categorical palettes, bivariate overlays), and see results instantly on a modern web map backed by Mapbox Vector Tiles.

Why LASEK

Current tools make it hard to (i) iterate quickly on styles, (ii) keep maps responsive at scale, and (iii) share reproducible styling choices. LASEK addresses these gaps by coupling LLM-assisted style generation with a scalable data-to-tiles pipeline and a lightweight web UI for rapid “try–tweak–compare” workflows.

Key Capabilities

  • LLM-guided styling: natural-language prompts propose styles (e.g., “graduated color ramp on population,” “categorize by continent,” “highlight top 10% by GDP”).
  • Real-time iteration: instant map updates, side-by-side comparisons, and saved presets.
  • Scale & performance: Spark/Scala backend for preprocessing; efficient rendering at interactive zoom levels.
  • Reproducibility: exportable style specs and dataset summaries for consistent sharing.

Artifacts: