The story
Ingesting the knowledge already circulating
Lionscraft built DocPythia to collect discussions from community communication streams and process them into candidate technical issues, solutions and documentation topics.
The software organises relevant conversations, identifies recurring questions and uses AI analysis to compare the information against the existing documentation set.
Instead of producing a generic summary of each conversation, the system looks for material that could improve published guidance: missing steps, changed behaviour, newly discovered fixes and explanations that maintainers may want to preserve.
A working demonstration showed 135 possible gaps
In a recorded product demonstration, the Lionscraft-built interface showed how community discussion records could be grouped and analysed against the NEAR documentation.
At the time of that demonstration, the application displayed 135 candidate documentation gaps. The reviewer could inspect the underlying issue and proposed change rather than blindly accepting AI-generated text.
That is the important feature of the number: it reflects a working detection and review process. It does not mean 135 production documentation defects were independently verified or all subsequently fixed.
From AI proposal to an editable document
The application includes a reviewer interface for examining an issue, comparing the suggested update with the documentation and editing the proposed text. Maintainers can inspect a preview, modify the language or reject the suggestion.
The system was designed as a human-in-the-loop workflow because technical operating instructions carry real consequences. It supports maintainers rather than silently publishing changes on their behalf.
Connecting the workflow to GitHub
Lionscraft built the software components needed to turn reviewed documentation changes into the repository workflow used by engineers.
The public DocPythia codebase contains the application interface, processing pipeline, data handling, AI analysis and GitHub pull-request service.
The implementation uses React and TypeScript for the interface, Express/TypeScript backend services, Prisma/PostgreSQL with pgvector, configurable AI analysis and integration with GitHub. A maintainer can therefore move from identified information to a proposed documentation change without manually rebuilding the entire context in a separate editor.
The public repository is MIT licensed.
Delivery and handover
The project was developed in collaboration with Meta Pool for the NEAR ecosystem. The historical project records include a signed collaboration agreement, an architecture diagram, implementation material and technical handover instructions.
Lionscraft's founder reports that DocPythia was delivered to Meta Pool for ongoing NEAR documentation maintenance. The public source code and demonstration establish the developed product; a separately verified production-usage record and measured time savings have not yet been established.