Building Enterprise RAG with RAGFlow

1. RAGFlow Architecture & Platform Core

• Understanding RAGFlow's core architecture: Task Executors, DeepDoc, Vector Store adapters, and Model Connectors
• Decoupling infrastructure: Connecting RAGFlow to external model servers (vLLM/TEI) and vector databases (OpenSearch/Infinity)
• Multi-tenancy, workspace separation, and dataset scoping within RAGFlow

2. DeepDoc Ingestion, Layout Analysis & Document Parsing

• Sourcing parser capabilities within RAGFlow: How DeepDoc handles multi-column layouts, nested tables, and OCR
• Document-centric chunking strategies: Naive, Book, Paper, Laws, Presentation, Picture, and Q&A templates
• Managing metadata, custom fields, and document status lifecycles inside RAGFlow Knowledge Bases

3. Hybrid Retrieval Configuration & Tuning

• Configure RAGFlow's hybrid retrieval pipeline using BM25 and dense-vector similarity
• Integrating cross-encoder reranking models within RAGFlow retrieval pipelines
• Visualizing retrieval hits, chunk scoring, and vector/lexical relevance weights
• Lab: RAGFlow Deployment, DeepDoc Ingestion & Retrieval Tuning
• Lab: Deploy RAGFlow using containerized orchestration connected to external LLM/embedding endpoints
• Lab: Ingest complex financial and clinical PDFs using DeepDoc across various chunking templates
• Lab: Compare candidate retrieval results while tuning vector/keyword weighting, similarity thresholds, top-k, and reranking

4. Enterprise Identity, Authorization & the RAGFlow Boundary

• Understanding RAGFlow's native security model: users, authentication, tenant/workspace permissions, dataset ownership, and knowledge-base sharing
• Integrating enterprise identity using OIDC/OAuth2 and mapping authenticated users into RAGFlow
• Separating authentication, platform permissions, dataset sharing, and retrieval authorization
• Using dataset boundaries and metadata filtering to enforce application-defined retrieval scope
• Extending RAGFlow with an application or API layer that translates enterprise roles and policy into permitted datasets and retrieval filters
• Authorization design patterns: native RAGFlow permissions, dataset-per-security-boundary, metadata-filtered retrieval, and application-controlled retrieval
• Understanding the limitations of each approach: administrative complexity, role/filter synchronization, filter correctness, policy drift, and large numbers of security boundaries
• Identifying when native RAGFlow controls are sufficient, when application-level augmentation is appropriate, and when authorization requirements justify a more custom architecture

5. Agent Workflows & Tool Integration

• Building visual workflow agents in RAGFlow: Knowledge Retrieval, Logic Nodes, and Categorization
• Extending agent capabilities: Connecting external web crawlers, Model Context Protocol (MCP) servers, and custom HTTP tools
• Balancing visual orchestration with application-owned business logic

6. Application Integration, Operations & APIs

• Interfacing with RAGFlow programmatically: Utilizing the REST API and Python SDK
• Injecting dynamic query filters and user context into API requests
• Session management, streaming responses, citation tracking, and telemetry monitoring
• Lab: RAGFlow Agents, Custom Tools & API Integration
• Lab: Build an multi-step agent inside RAGFlow that combines Knowledge Base retrieval with an external API tool
• Lab: Integrate RAGFlow programmatically into an external application using the Python SDK / REST API
• Lab: Inject dynamic runtime metadata filters via API calls and capture structured responses with source citations