-
Notifications
You must be signed in to change notification settings - Fork 852
Open
Description
I attempted to generate wikis using the Qwen/Qwen3-Embedding-8B Embedding model and the deepseek-ai/DeepSeek-R1 general model. Based on the results, the following issues were identified:
- The analysis of page content and system architecture is superficial, with limited practical value. There is significant room for improvement in analytical depth.
- High overlap exists across content in different wiki pages, and some hyperlink addresses are incorrect.
For the points mentioned above, I have several optimization ideas, but I'm not sure if they are feasible:
- Extract the deepwiki capabilities into an MCP plugin and leverage Cursor’s existing Agent capabilities to strengthen the analytical outcomes.
- Utilize the project’s historical architecture documentation knowledge base through RAG (Retrieval-Augmented Generation) to enhance the depth of analysis.
- If holistic project analysis faces bottlenecks, can we break through from single points? Improve the architectural analysis depth for a specific directory within the project.
sng-asyncfunc
Metadata
Metadata
Assignees
Labels
No labels