paperlab

paperlab is a multiagent system to help students to understand papers in machine learning and deep learning.

Work flow

flowchart LR
    Acquirer((acquirer)) --> Sources[PDF / Git repo]
    Sources --> Dissector((dissector))
    Dissector --> Spec[spec.md]

    Spec --> Implementer((implementer))
    Sources --> Implementer
    Implementer --> CodeMap[code_map.md]
    Implementer --> DeepDive[code_map__slug__component.md]

    Spec --> Critic((critic))
    CodeMap --> Critic
    Critic --> Review[critic_reviews.md]

    Spec --> Tutor((tutor))
    Tutor -. invokes .-> Explainer((explainer))
    Explainer --> ConceptBound[concept-slug.md]
    Tutor --> Concept[concept.md]
    Tutor --> Synth[synth__a__b.md]
    Tutor --> TutorLog[tutor_log.md / tutor_notes.md]

Loading

The user-facing concept interface is the tutor (/tutor <slug>); it invokes the explainer in the background. The experimenter suite (multi-paper empirical comparison) runs alongside this per-paper flow: the comparator (shipped) compares methods across papers along an axis and writes comparison.md to <vault>/experiments/<topic>/; the experimenter, coder, and evaluator are designed (see ROADMAP.md).