zjuncher/xiaoyan-ai-dev-assistant

This is an AI-powered knowledge base and question-answering assistant designed for software development teams. The core problem it solves is information overload in team settings: developers, new hires, and project managers can upload documents such as PDFs, Word files, Markdown docs, and plain text, then ask questions in natural language and receive accurate, context-aware answers instead of hunting through folders or asking colleagues repeatedly. The system uses a technique called RAG, short for Retrieval-Augmented Generation, which means it finds the most relevant chunks of stored documents before asking an AI model to generate an answer. It combines two search strategies: semantic vector search using Pinecone, a cloud vector database, and keyword search using BM25, a traditional text-ranking algorithm. Results from both are merged and re-ranked for quality. The AI model used for generating answers is Qwen-Plus, and conversations are streamed back to the browser in real time. Memory works in two layers: short-term memory stores the current conversation in Redis, a fast in-memory database, and is automatically summarized when it grows too large. Long-term memory lets users save personal preferences or important facts that persist across sessions. A developer or team lead would use this when their team accumulates a lot of internal documentation, coding standards, onboarding materials, or FAQ documents and needs a searchable, conversational interface to access all of it instantly. The tech stack is Spring Boot 3.5 with Java, LangChain4j for AI integration, MySQL for storage, Redis for caching, and a lightweight Vue 3 frontend.