{% extends "base.html" %} {% block title %}AI review{% endblock %} {% block content %}

AI review

Upload a draft PDF. The top-N AI reviewers most relevant to your topic are selected from the active database, each produces structured critique, and the results are clustered and ranked by commonality × importance.

{% if not any_provider_configured %}
No LLM API is configured. Head to the Model page to pick a provider and set its key, or edit config.yaml directly — then come back here.
{% endif %}
{% if llm_status %}

Currently using {{ llm_status.review_provider }} / {{ llm_status.review_model }}. Change on the Model page.

{% endif %}

1. Choose a reviewer database

Each database is a pool of reviewer personas tailored to a specific field. Two databases are bundled by default: Computer Architecture and Machine Learning & AI. To build a new database for a different field, see Database.

2. Number of reviewers

How many reviewer personas from the database should review this paper. Each selected reviewer runs in parallel via one LLM call and produces 5–10 review comments. Recommended: {{ recommended_min_n_reviewers }}–{{ recommended_max_n_reviewers }} — this balances coverage and cost. Below {{ recommended_min_n_reviewers }} loses the cross-reviewer consensus signal that the downstream clustering relies on; above {{ recommended_max_n_reviewers }} adds little new signal but scales LLM cost and wall-time linearly. Default is {{ default_n_reviewers }}; hard range is {{ min_n_reviewers }}–{{ max_n_reviewers }}.

3. Upload the PDF

Typical end-to-end time is 1–5 minutes, depending on paper length, API latency, and number of reviewers.

Recent reviews

{% if recent_jobs %} {% for j in recent_jobs %} {% endfor %}
RunPDF fileStatusStarted
{{ j.id_short }} {{ j.filename or '—' }} {{ j.status }} {{ j.started_at[:19].replace('T', ' ') if j.started_at else '' }} {% if j.status == 'done' %} View {% else %} Status {% endif %}
{% else %}

No reviews yet.

{% endif %}
{% endblock %}