How xinyuez/chat-your-doc works
Main Components
• Experimental App Tiers – Separate the applications into three main layers: Simple, Intermediate, and Advanced. Each tier includes streamlit-based demos and standalone scripts.
• LLM Agent/Workflow Modules – Represent various agentic flows, such as LangChain agents, LlamaIndex workflows, CrewAI agent systems, and LangGraph processes. These modules form the core logic of how the system “thinks” and interacts with provided data.
• Data Processing & Retrieval – Include components for vectorization and storage (using FAISS, Chroma etc.), RAG engines, and document ingestion pipelines.
• External Integrations – External services such as OpenAI’s API, Ollama, computer vision modules (e.g., GroundingDINO, YOLOv8), and audio/music synthesis components.
• Notebooks & Experimentation Environment – Represent the notebooks directory as a dynamic space for rapid experiments, testing, and prototyping.
• Assets & Tools – Include repositories of configuration (environment.yml, requirements.txt), images, and deployment scripts (like Dockerfile in the wine-price-app).
Relationships and Data Flow
• Experimental & Notebook Environment: Represent the notebooks folder as a sandbox that interacts similarly with the agents and storage layer, reinforcing iteration and experimentation.
Architectural Patterns
• Modular and Layered Architecture: Emphasize separation between UI, application logic (agentic processing), data management, and external service integrations.
• Microservices-Like Structure for Apps: Note that while the project is monorepo-based, its sub-app divisions (simple, intermediate, advanced) can be conceptually treated as separate micro-applications that share core underlying technology stacks.
Inclusion of Detailed Elements
• Annotations: For components with complex interactions (such as the multi-agent workflow or the adaptive RAG patterns), include brief notes or annotations that explain their role (e.g., “Handles multi-query augmentation” or “Integrates retrieval and summarization”).