Chat-centric scientific assistant — from keyword search to RAG-powered writing, orchestrated by LangGraph.
ChatGPT-style interface with SSE streaming, knowledge base selection, and multiple tool modes (QA, citation lookup, review outline, gap analysis).
LangChain abstraction supporting OpenAI, Anthropic, Aliyun, Volcengine, Ollama, and mock providers with per-session switching.
LlamaIndex-powered with ChromaDB vector store, GPU-aware HuggingFace embeddings, hybrid retrieval, and cited answers.
Orchestrated search → dedup → HITL → crawl → OCR → index workflow with interrupt/resume and checkpointing.
Federated search across Semantic Scholar, OpenAlex, arXiv, and Crossref with standardized metadata.
Three-stage pipeline with Git-style HITL conflict resolution — DOI hard dedup, title similarity, LLM verification.
Persistent subscription rules with CRUD API, frequency scheduling, and automatic incremental updates.
Multi-channel PDF download via Unpaywall/arXiv, native text extraction with PaddleOCR GPU fallback.
Summarization, citation generation (GB/T 7714, APA, MLA), review outlines, and gap analysis.
Model Context Protocol server exposing tools, resources, and prompts for Claude Code, Codex, and other AI IDEs.
Full Chinese/English bilingual interface with react-i18next and automatic language detection.
shadcn/ui + Radix components, Framer Motion animations, icon sidebar, and responsive design with TailwindCSS v4.