Skip to content

README.md

Latest commit

 

History

History
28 lines (18 loc) · 948 Bytes

README.md

File metadata and controls

28 lines (18 loc) · 948 Bytes
Raw

Before / After Benchmark

This example shows the core claim of AgentFigureGallery with the same tiny synthetic embedding dataset:

before: prompt-only plotting
after: agent query -> visual references -> human preference -> selected bundle -> plotting code

Run:

python examples/before_after_benchmark/render_before_after.py

Outputs:

  • examples/before_after_benchmark/figures/before_after_benchmark.png
  • examples/before_after_benchmark/figures/before_after_benchmark.pdf
  • examples/before_after_benchmark/figures/before_after_benchmark.svg

Preview

Before after benchmark

Why This Matters

The left panel represents a typical prompt-only first pass: visible data, but weak hierarchy and no reference taste. The right panel applies the same design logic as examples/generated_embedding_plot, using selected references to produce a cleaner publication-style result.