STAR 项目陈述模板（中英文）¶

S（背景）：
在 [业务/课程/科研/实习] 场景下，我们遇到 [核心问题]。

T（任务）：
我的任务是负责 [模块/系统/优化目标]，目标是把 [指标] 从 [原始值] 提升到 [目标值]。

A（行动）：
我做了 3 件关键事情：
1. [关键动作 1]
2. [关键动作 2]
3. [关键动作 3]

R（结果）：
最终 [指标 1] 提升到 [结果]，[指标 2] 降低到 [结果]，
并且 [上线/演示/复盘/业务收益]。

Situation:
In [business/research/project] context, we had a problem with [core issue].

Task:
I was responsible for [module/system/optimization goal], and the target was to improve [metric] from [baseline] to [target].

Action:
I took three key actions:
1. [action 1]
2. [action 2]
3. [action 3]

Result:
As a result, [metric 1] improved to [result], [metric 2] dropped to [result],
and the system was [deployed/validated/demonstrated].

在企业知识库问答场景中，原系统经常答非所问，且延迟较高。
我的任务是负责检索链路和评测体系，把 Faithfulness 和延迟同时优化。
我主要做了三件事：第一，重构分块和混合检索；第二，引入 reranker 和缓存；
第三，建立 RAGAS + 人工抽检评测流程。
最终 Faithfulness 从 0.62 提升到 0.87，P99 从 5.4s 降到 2.1s，并形成了可复用的上线流程。

In an enterprise knowledge-base QA scenario, the original system often produced irrelevant answers and had high latency.
I was responsible for the retrieval pipeline and evaluation workflow, with the goal of improving both faithfulness and latency.
I made three key changes: redesigned chunking and hybrid retrieval, introduced reranking and caching, and built an evaluation pipeline with RAGAS plus manual review.
As a result, faithfulness improved from 0.62 to 0.87, and P99 latency dropped from 5.4s to 2.1s. We also turned the workflow into a reusable release process.