LLM Wiki
An industrial-grade LLM knowledge base system developed on the DeepSeek Harness framework. It supports a full-pipeline set of capabilities including document collection and compilation, knowledge graph construction, traceable question answering, and online evaluation and optimization.
Timeline (last 90 days)
Andrej Karpathy proposed the LLM Wiki pattern, which uses Markdown pages plus YAML metadata to organize knowledge, allowing humans and AI agents to maintain it together and query it on demand.
The core design of LLM Wiki includes three components: preserved raw files, interlinked note pages, and a specification document that guides the AI agent in maintaining the wiki.
This LLM Wiki supports incremental document compilation, allowing newly added materials to be ingested incrementally without requiring a full rebuild
The core features of this LLM Wiki include a visual workbench, document compilation and incremental updates, a dual-form knowledge graph, traceable Q&A, online evaluation and optimization, and a system settings page
Traceable Q&A can link any response back to the original knowledge fragment
The entire LLM Wiki is broken down into 11 development phases, proceeding in the order of environment setup, project foundation development, core functional module development, and feature acceptance
This LLM Wiki has officially launched on the team's in-house DeepSeek Harness Studio desktop application
The features implemented in this LLM Wiki project include: a workbench Dashboard, document management compilation and upload, incremental document compilation, a dual-form knowledge graph, traceable knowledge base Q&A, and online evaluation and optimization.
The LLM Wiki project was broken down into 11 development phases following the main thread of environment setup → project foundation development → core feature module development → feature acceptance.
An LLM Wiki is a type of knowledge base whose core function is to collect and compile scattered materials into structured knowledge and then support question answering through retrieval augmentation.
2 more timeline events
All Facts (12)
Andrej Karpathy proposed the LLM Wiki pattern, which uses Markdown pages plus YAML metadata to organize knowledge, allowing humans and AI agents to maintain it together and query it on demand.
50%UnverifiedThe core design of LLM Wiki includes three components: preserved raw files, interlinked note pages, and a specification document that guides the AI agent in maintaining the wiki.
50%UnverifiedThis LLM Wiki supports incremental document compilation, allowing newly added materials to be ingested incrementally without requiring a full rebuild
50%UnverifiedThe core features of this LLM Wiki include a visual workbench, document compilation and incremental updates, a dual-form knowledge graph, traceable Q&A, online evaluation and optimization, and a system settings page
50%UnverifiedTraceable Q&A can link any response back to the original knowledge fragment
50%UnverifiedThe entire LLM Wiki is broken down into 11 development phases, proceeding in the order of environment setup, project foundation development, core functional module development, and feature acceptance
50%UnverifiedAn LLM Wiki is a type of knowledge base whose core function is to collect and compile scattered materials into structured knowledge and then support question answering through retrieval augmentation.
50%UnverifiedThe features implemented in this LLM Wiki project include: a workbench Dashboard, document management compilation and upload, incremental document compilation, a dual-form knowledge graph, traceable knowledge base Q&A, and online evaluation and optimization.
50%UnverifiedThe LLM Wiki project was broken down into 11 development phases following the main thread of environment setup → project foundation development → core feature module development → feature acceptance.
50%UnverifiedThe design inspiration for claude-obsidian comes from the LLM Wiki pattern proposed by Andrej Karpathy
50%UnverifiedThe core idea of the LLM Wiki pattern is to have large language models act as knowledge base administrators, automatically handling the understanding, linking, and archiving of content to form a self-growing knowledge network
50%UnverifiedThis LLM Wiki has officially launched on the team's in-house DeepSeek Harness Studio desktop application
50%