ingar30/reviewer

Reviewer is a Python-based tool that automates the peer-review process for academic economics papers using multiple AI agents working together. Think of it as an automated research assistant that reads a paper (provided as a PDF), runs it through several specialized reviewers, each checking a different aspect of the work, and then produces a single consolidated review report. The workflow starts by converting the PDF into structured text and checking that the conversion was clean enough to trust. From there, a set of mandatory reviewers always run: one checks internal cross-references and numbering, another verifies the bibliography, and a third catches grammar and copyediting issues. A pool of optional reviewers is also selected dynamically based on the paper, covering things like numerical accuracy, how well evidence supports claims, literature coverage, identification strategy, and robustness of results. Each reviewer produces a structured output in a standardized format, which is then validated, deduplicated, and assembled into a final Markdown report. The project uses Python 3.12 and a tool called Codex CLI (an AI coding and agent runner) to orchestrate the agents. It is designed for reproducibility: the shared code handles prompts, schemas, scripts, and tests, while the actual papers and generated reviews stay private on the user's own machine. Researchers or teams doing systematic paper evaluation would use this to get fast, consistent, multi-angle feedback on draft economics papers without manually coordinating multiple review passes.