借助 Query-Driven Observability 将 Agentforce AI 调试时间从两周缩短至当天

In our Engineering Energizers Q&A series, we highlight the engineering minds driving innovation across Salesforce. Today, we spotlight Kishore Chaganti, Principal Software Engineer on the Interactive Data Science team, who scales the Einstein Notebooking platform to provide secure, production-grade AI debugging for Data 360 and Agentforce. This empowers engineers to run investigations across more than 60 Agentforce features, with 600 users, and massive datasets exceeding 400 million records and 800 GB of data.

在我们的 Engineering Energizers Q&A 系列中，我们突出了推动 Salesforce 创新的工程头脑。今天，我们聚焦 Kishore Chaganti，Interactive Data Science 团队的 Principal Software Engineer，他扩展了 Einstein Notebooking 平台，为 Data 360 和 Agentforce 提供安全、生产级的 AI 调试。这使工程师能够针对超过 60 个 Agentforce 功能、600 名用户以及超过 4 亿条记录和 800 GB 数据 的海量数据集进行调查。

Explore how the team solved the challenge of opaque AI agent behavior by providing query-driven access to production data, slashing investigation times from two weeks to a single day through Spark-based workflows that expose document chunks, embeddings, and session-level feedback across Data 360 systems.

探索团队如何通过提供查询驱动的生产数据访问来解决不透明的 AI 代理行为挑战，将调查时间从两周缩短到一天，通过基于 Spark 的工作流，这些工作流暴露了 Data 360 系统中的文档块、嵌入和会话级反馈。

What is your team’s mission enabling production Agentforce AI debugging through Einstein Notebooking?

您的团队通过 Einstein Notebooking 启用生产 Agentforce AI 调试的使命是什么？

We build systems that enable engineers and data scientists to debug AI agents against actual production behavior. This approach allows teams to identify, reproduce, and analyze issues using real data and system interactions.

我们构建系统，使工程师和数据科学家能够针对实际生产行为调试 AI agents。此方法允许团队使用真实数据和系统交互来识别、重现和分析问题。

To make this possible, the team develops unified workflows that bridge the gap between data access, debugging, and validation. These tools empower data scientists and engineers to investigate production scenarios directly instead of relying on incomplete staging environments.

为了实现这一点，团队开发了统一的工作流，弥合数据访问、调试和验证之间的差距。这些工具使数据科学家和工程师能够直接调查生产场景，而不是依赖不完整的暂存环境。

Such work requires full vi...