Academic Personal AI Infrastructure — an end-to-end research assistant pipeline covering paper discovery, LLM analysis, scholar tracking, Paper2Code reproduction, and multi-agent collaboration.
PaperBot is an end-to-end academic AI infrastructure for researchers, consolidating paper aggregation, quality assessment, knowledge management, code reproduction, scholar monitoring, and conference tracking into a single system. It offers three interaction interfaces: a Web dashboard (Next.js), a terminal CLI (Ink), and a backend API (FastAPI + SSE), with long-running tasks driven by the ARQ async task queue.
For paper discovery, PaperBot aggregates data from arXiv, Semantic Scholar, OpenAlex, HuggingFace Daily Papers, and papers.cool with cross-query deduplication and scoring. The DailyPaper feature auto-generates daily reports with LLM-enhanced summaries and trend analysis, pushing to 7 channels including Email, Slack, and Telegram. The LLM-as-Judge module performs multi-round calibrated scoring across 5 dimensions (Relevance, Novelty, Rigor, Impact, Clarity) to automatically filter low-quality papers. Deadline Radar tracks conference deadlines with CCF ranking and research direction matching.
For knowledge management, Paper Library supports collection and multi-format export (BibTeX/RIS/Markdown/CSL-JSON/Zotero sync). Structured Cards auto-extract methods, datasets, conclusions, and limitations via LLM. Related Work generates review drafts from stored papers. The Memory System builds research memory with FTS5 + BM25 full-text search, accompanied by the MemoryBench evaluation suite achieving 0.873 Recall@5, aligned with LongMemEval, LoCoMo, and other academic benchmarks.
For reproduction and collaboration, Paper2Code converts papers to code skeletons (Planning → Analysis → Generation → Verification) with self-healing debugging. AgentSwarm provides multi-agent orchestration integrated with Claude Code, supporting Docker/E2B sandbox execution. Scholar Tracking enables multi-agent scholar monitoring with PIS influence scoring. Deep Review simulates the full peer review pipeline.
The backend is built with Python + FastAPI, database managed via SQLAlchemy + Alembic, with code quality enforced by Pyright, Black + isort, and SonarQube. Core modules (Topic Search, DailyPaper, LLM-as-Judge, Push/Notify, Deadline Radar, Paper Library) are Production-grade, while Paper2Code, AgentSwarm, and Memory are Usable-grade.
Note: The project is actively developed on the
devbranch with no formal Release; the Vercel preview deployment link's long-term availability is uncertain; the full list of supported LLM providers is to be confirmed; no Docker Compose deployment is available yet.