Custom content source best practices（自定义内容源最佳实践）¶
To improve AIP Assist responses, it is vital to understand retrieval augmented generation ↗ (RAG), the underlying mechanism used. This process involves dividing your content into concise, focused paragraphs. When a question is asked, RAG identifies and retrieves the paragraphs most pertinent to the question. AIP Assist then uses these selected paragraphs, along with information from public documentation to formulate a response.
Given this approach, the granularity of your content can significantly impact the quality of responses. The best practice is to organize your content into discrete sections based on headings and subheadings. Each heading should ideally be dedicated to a single topic or question, ensuring that the content under it is highly relevant and specific to that subject.
Structure for clarity¶
Consider a section in your content discussing various features of a software product. Instead of creating a long, multi-topic paragraph like this:
Feature A helps with productivity by automating tasks. Feature B enhances security through encryption. Feature C offers real-time collaboration tools, and Feature D provides detailed analytics.
Break it down into focused sub-sections:
## Feature A: Automation
Feature A boosts productivity by automating repetitive tasks, streamlining workflows.
## Feature B: Security
Feature B secures data with advanced encryption techniques, protecting against unauthorized access.
## Feature C: Collaboration
Feature C enables real-time collaboration, allowing teams to work together seamlessly.
## Feature D: Analytics
Feature D offers comprehensive analytics, giving insights into performance metrics.
Structuring your content this way allows AIP Assist to more efficiently retrieve relevant information in response to specific inquiries, improving the accuracy and helpfulness of its answers.
Avoid collisions with public documentation¶
AIP Assist also searches through the Palantir platform's public documentation to craft answers. Therefore, it is crucial to ensure that your custom content does not unintentionally overlap or collide with the information found in these public documents. When using terms that are common or similar to those found in public documentation (for example, "build," "code repository"), it is important to clearly define or disambiguate them within your context. This clarification helps AIP Assist to distinguish between similarly named concepts and provide more accurate, context-specific responses.
Avoid deprecations or overrides¶
When updates to processes or workflows occur, avoid marking obsolete documents as deprecated or adding overrides. Instead, update the content at its source or remove any invalid documents.
For example, rather than using the following:
(Note: This document is now deprecated.)
In case of a high priority issue, contact abx@xyz.com directly.
or
In case of a high priority issue, contact abx@xyz.com directly.
...
...
(Update: As of July 31, create a service ticket on the management system instead of sending an email to abx@xyz.com.)
Update the content to a consolidated and current form, such as:
In case of a high priority issue, create a service ticket on the management system.
中文翻译¶
自定义内容源最佳实践¶
为提升 AIP Assist 的响应质量，理解其底层机制——检索增强生成 ↗（RAG，retrieval augmented generation）至关重要。该过程将您的内容拆分为简洁、聚焦的段落。当用户提出问题时，RAG 会识别并检索与问题最相关的段落。随后，AIP Assist 会利用这些选中的段落，结合公共文档中的信息来生成回答。
基于这一机制，内容的粒度会显著影响回答质量。最佳实践是根据标题和子标题将内容组织成独立的章节。每个标题应尽量专注于单一主题或问题，确保其下的内容与该主题高度相关且具体。
清晰的结构化¶
假设您的内容中有一个章节讨论了某款软件产品的多项功能。请避免编写类似下面这种冗长、涵盖多个主题的段落：
功能 A 通过自动化任务提升生产力。功能 B 通过加密增强安全性。功能 C 提供实时协作工具，功能 D 则提供详细的分析功能。
请将其拆分为聚焦的子章节：
## 功能 A：自动化
功能 A 通过自动化重复性任务提升生产力，简化工作流程。
## 功能 B：安全性
功能 B 采用先进的加密技术保护数据，防止未经授权的访问。
## 功能 C：协作
功能 C 支持实时协作，使团队能够无缝协同工作。
## 功能 D：分析
功能 D 提供全面的分析功能，帮助洞察性能指标。
以这种方式组织内容，AIP Assist 能够更高效地检索与特定查询相关的信息，从而提高回答的准确性和实用性。
避免与公共文档冲突¶
AIP Assist 还会搜索 Palantir 平台的公共文档来生成答案。因此，务必确保您的自定义内容不会与这些公共文档中的信息意外重叠或冲突。当使用与公共文档中常见或相似的术语时（例如“构建”、“代码仓库”），请务必在您的上下文中明确定义或区分这些术语。这种澄清有助于 AIP Assist 区分名称相似的概念，并提供更准确、更具上下文针对性的回答。
避免弃用或覆盖¶
当流程或工作流发生更新时，请避免将过时文档标记为“已弃用”或添加覆盖说明。相反，应直接更新源内容或删除无效文档。
例如，不要使用以下写法：
（注意：本文档现已弃用。）
如遇高优先级问题，请直接联系 abx@xyz.com。
或
如遇高优先级问题，请直接联系 abx@xyz.com。
...
...
（更新：自 7 月 31 日起，请在管理系统上创建服务工单，而非发送邮件至 abx@xyz.com。）
请将内容更新为整合后的当前版本，例如：
如遇高优先级问题，请在管理系统上创建服务工单。