What they built

ReviewerZero Academic Suite

The reviewer-zero project is an advanced, automated academic peer-review system built on the NitroStack framework. Its primary objective is to streamline the peer-review process by utilizing a multi-agent architecture where specialized AI tools handle distinct aspects of paper analysis, evaluation, and synthesis. At the core of the backend architecture are three main components: ReviewerModule: This houses a suite of 16+ specialized AI agents designed for intricate peer-review tasks. Key agents include the ClaimsAuditorTool (for verifying logical soundness), RigorAgentTool (for assessing methodology), CitationScoutTool (for checking reference validity), and WeaknessDetectorTool. These agents are powered by an LLMService (integrating with models via Groq/Gemini), a SemanticScholarService for academic lookups, and a LatexParserService for processing mathematical formatting. Aegis Interceptor: To ensure stability and resilience, the AegisInterceptor acts as a self-healing middleware during tool executions. It catches failures—such as API rate limits or network errors from LLM providers—and automatically attempts retries using exponential backoff. It meticulously logs all these healing events so users can monitor system resilience. OrchestratorModule: This module acts as the "manager" of the system. Its OrchestratorTool coordinates the workflow of the various review agents, taking a paper as input, delegating tasks, and synthesizing their independent findings into a cohesive, structured ReviewPayload. This ultimately produces a comprehensive ReviewLetter and a final grading of the paper. On the frontend, the project features a modern, interactive web dashboard built with Next.js (src/widgets/app/review-dashboard). This dashboard serves as the user-facing interface, allowing researchers to visualize the real-time execution of the AI agents, monitor system health (such as Aegis healing events), and read the synthesized peer reviews. The UI is designed with a premium