magic-meet/retrieve-pro-plus

This system handles uploading, parsing, storing, and searching academic PDF papers. The README is written in Chinese, the description below is drawn from that documentation. When a user uploads a PDF, the system breaks it into chunks and sends those pieces through a message queue for asynchronous processing. A Python-based parser using a tool called GROBID reads the PDF and extracts structured information about the paper, including its title, authors, sections, paragraphs, and references. That structured data is saved to a MySQL database. From there, two parallel processes prepare the content for search. A change-tracking component called Canal monitors the database for new or updated records and pushes those changes into Elasticsearch, a search engine that enables fast text matching. A separate Python service generates a numerical vector representation of each paper using an AI language model called Qwen3 Embedding, and that vector is also stored in Elasticsearch. This setup supports three types of search. Keyword search finds papers whose text contains matching words. Semantic search converts a query into a vector and finds papers with similar meaning even if the exact words differ. Hybrid search combines both approaches. The frontend is a web interface built with Vue 3 where users can upload papers, browse the list, view details, and run searches. The entire system is packaged as a Docker Compose configuration, so all components, including the database, message queue, search engine, file storage, and AI model service, start with a single command. The system requires at least 8 GB of RAM, and 16 GB or more is recommended when running all components simultaneously, since the AI embedding model is memory-intensive.