Build coaching
HR team at a large Korean company
We coached the HR team as they built their own in-house agent to score written answers from job applicants. The agent calls Claude Sonnet through Amazon Bedrock.
The problem: the expected cost was above the internal budget, and an early accuracy figure turned out to count parsing failures as zero scores.
What changed: the goal moved from a final score to first-pass material sorted into three bands (high, human review, low). Re-scoring calls were removed and prompt caching was added to lower cost.
국내 대기업 HR 팀이 사내 채용 서술형 채점 에이전트를 직접 만들도록 코칭했습니다. 코드는 고객사 소유입니다.