AdamPlatin123/Open-Deep-Research-workflow-on-Dify

What it solves

This project provides a workflow implementation for the Dify platform that replicates the "Deep Research" capability. It allows users to generate comprehensive, structured research reports of up to 10,000 words within five minutes by automating the process of topic decomposition, multi-source information retrieval, and content synthesis.

How it works

The system uses a modular design that integrates multiple models and data sources. It begins by using Gemini 2.0 Flash to decompose a user's topic into multiple dimensions for deep analysis. It then employs a hybrid retrieval engine that combines local knowledge bases with web search APIs (Wikipedia, Google, and Bing). A cascaded model architecture (2>1) uses conditional branches and conversation turn markers to control the processing pace. Finally, models like deepseek-r1-distill are used to generate structured Markdown reports on a paragraph level.

Who it’s for

Researchers and users of the Dify platform who need to automate the production of long-form, structured research reports based on both internal data and live web information.

Highlights

• Multi-source retrieval: Combines local knowledge bases with Google, Bing, and Wikipedia APIs.
• Intelligent topic parsing: Uses multi-level decomposition to analyze topics across four dimensions.
• Modular architecture: Supports the flexible replacement of underlying LLMs and data sources.
• High-efficiency generation: Capable of producing extensive structured reports quickly using a cascaded model approach.