跪拜 Guibai
← All articles
Frontend · JavaScript

One Developer's Answer to Poetry Pop-Ups: An 800,000-Poem, No-Ad Archive

By 夕水 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

A full-featured archive is buildable with open-source libraries alone, but keeping 800,000 entries accurate requires either API budgets or a volunteer proofread crew; this effort chooses the crowd.

Summary

Pop-up ads, VIP paywalls, and unreliable servers drove the decision to rent a server and rebuild an archive from scratch. The result is a public-interest site holding nearly 800,000 poems from ancient to contemporary China, with pinyin annotations, text-to-speech, and one-click simplified/traditional switching. No ads, no charges, and a side page for original poems.

Content, not code, is the time sink. Local open-source libraries like pypinyin batch-process pronunciation, and scripts handle simplified/traditional conversion, but annotations and proofreading are done by hand because continuously calling large model APIs is too expensive. With 800,000 poems, that adds up to a massive manual burden. An online correction button now sits on the site, and anyone who spots a typo or pinyin error can click it; contributor names appear on the homepage. The code is closed for now, and the URL is deliberately not posted here to avoid being flagged as traffic bait.

Takeaways
— The archive holds nearly 800,000 poems spanning ancient to contemporary times, including classical Chinese and some modern poetry.
— Pinyin is batch-processed using the local open-source library pypinyin; simplified/traditional conversion is script-based.
— Manual annotation and correction are required because large-model API calls are unaffordable for a non-commercial project.
— An online correction button accepts anonymous or nicknamed submissions, and contributor names are listed on the homepage.
— The code is not open-source for now because both the backend and frontend were rewritten.
— The site is ad-free and free of charge, described as a purely public-interest project.
Conclusions

AI-assisted IDEs like Trae lower the barrier to building a full-featured site, but the manual proofreading of 800,000 poems shows that content, not code, remains the bottleneck.

Closed source plus community corrections creates a single-maintainer dependency: if updates stop, the volunteer effort has nothing to anchor to.

Concepts & terms
Trae
ByteDance's AI-assisted IDE that generates and refactors code through natural language prompts, used here to rebuild the poetry site.
From the discussion

The discussion pivots on the archive's legal standing and discoverability. One reader flags the public security bureau filing requirement and recounts a rejection when self-listing as both contact and organization; the owner responds that the filing is in progress. A separate request for the site address is answered directly, while another reader offers only aesthetic approval.

— The site likely requires Chinese public security bureau filing, and listing oneself as both emergency contact and affiliation can lead to rejection.
— The archive's address is not immediately visible on the page, but the owner supplies it on request.
— The site's visual design earns approval, though that praise stands apart from the technical and compliance concerns.
— A tension exists between the compliance burden (filing) and the owner's apparent solo operation, as the filing rejections example illustrates.
Featured comments
骑猪走天下

This needs to be filed with the public security bureau, right? When I filed before, I put myself as both the emergency contact and the organization, and it was directly rejected [sweat]

夕水

I'm already applying [a polite smile]

我不想上班

Didn't see the address

夕水

Search Xishui Literature Network on Baidu, or visit aspoem.eveningwater.com

Tim_Zhou

The website style is the type I like

See top comments, translated →
Source: juejin.cn ↗ Google Translate ↗ Backup ↗