Remy v1 — typed reading-group copilot

Generate six typed cards from a paper excerpt and audit the analogy for groundedness against the source. Part of the À la carte reading-group toolkit.

Architectural lineage Remy v1 adapts two recent ideas from the agent-systems literature:

  • Code World Models for General Game Playing (Lehrach et al., 2025) — typed code components refined via tree search and unit-test verification. The six recipe-card types are the reading-group analog. arXiv:2510.04542
  • AutoHarness (Tang et al., 2024) — code-as-harness pattern where the LLM produces typed action verifiers. The analogy-groundedness rule is the small-scale analog of is_legal_action. Paper
Paper
Audience level