展示HN：我在Gemini中仅使用系统提示实现了“二次生成”功能
嗨，HN，
我是一名“无代码”架构师。根据Reddit上一位RAG工程师的技术建议，我在Gemini 3.0 Pro中实现了“二次生成”（将事实提取与创作分开），仅使用系统提示。
它通过严格将“获取”阶段与“写作”阶段分开，从结构上消除了幻觉。
GitHub（提示）：
[https://github.com/dosanko-tousan/Gemini-Abhidhamma-Alignment](https://github.com/dosanko-tousan/Gemini-Abhidhamma-Alignment)
完整故事（Medium）：
[https://medium.com/@dosanko_tousan/the-engineer-and-the-buddhist-practitioner-how-a-reddit-comment-fixed-my-ai-architecture-f07957e03760](https://medium.com/@dosanko_tousan/the-engineer-and-the-buddhist-practitioner-how-a-reddit-comment-fixed-my-ai-architecture-f07957e03760)
我很想听听你们对这种“逻辑绑定”架构的看法。
查看原文
Hi HN,<p>I am a "No-Code" architect.
Based on technical advice from a RAG engineer on Reddit, I implemented "Two-Pass Generation" (separating fact extraction from composition) in Gemini 3.0 Pro using only System Prompts.<p>It works by strictly separating the "fetching" phase from the "writing" phase, structurally eliminating hallucinations.<p>GitHub (Prompt):
<a href="https://github.com/dosanko-tousan/Gemini-Abhidhamma-Alignment" rel="nofollow">https://github.com/dosanko-tousan/Gemini-Abhidhamma-Alignmen...</a><p>Full Story (Medium):
<a href="https://medium.com/@dosanko_tousan/the-engineer-and-the-buddhist-practitioner-how-a-reddit-comment-fixed-my-ai-architecture-f07957e03760" rel="nofollow">https://medium.com/@dosanko_tousan/the-engineer-and-the-budd...</a><p>I'd love to hear your thoughts on this "Logic-Bonded" architecture.