Skip to content

Latest commit

 

History

History
52 lines (42 loc) · 1.97 KB

File metadata and controls

52 lines (42 loc) · 1.97 KB

User as Code — interactive paper companion

A React website for the paper User as Code: Executable Memory for Personalized Agents. It explains the mechanism and lets you interactively explore the cases bundled for three reported evaluations (LOCOMO, LongMemEval, and Analytical Inference) plus an explicitly exploratory alert protocol. The LOCOMO explorer currently contains the legacy 600-question subset, not the full 1,986-question suites.

Stack

  • Vite + React 18
  • Tailwind CSS
  • Recharts (interactive charts)
  • Framer Motion (animation)

Develop

npm install
npm run dev        # http://localhost:5173

Build (static)

npm run build      # → dist/  (fully static, deployable anywhere)
npm run preview

Data

The site is driven by compact JSON bundles in public/data/, generated from the raw experiment results by:

python3 ../scripts/build_site_data.py

This reads experiments/results/*.json, the source benchmarks, and the active-service scenario definitions, and writes:

file contents
locomo.json Legacy LOCOMO subset: the first 60 QAs from each of 10 conversations (600 total; not random) × 7 systems, with conversation-evidence context and dual-judge grades
longmemeval.json 500 LongMemEval QAs × 7 systems, dual-judge grades
analytical.json 100 analytical cases × 5 systems, with executed Python traces
active.json 60 exploratory alert scenarios with multi-session context; not publication evidence
summary.json aggregate tables / chart data supported by the current paper

Re-run the script if the underlying results change, then rebuild.

Each explorer case shows

  • (a) the benchmark result / gold answer
  • (b) the context (conversation evidence, seeded sessions, or record set)
  • (c) each system's response (including UaC's executed code trace)
  • (d) the stored grading fields where applicable; legacy cross-judge data are not current headline evidence