Job
청킹 전략, 임베딩 선택, 벡터 인덱싱, 밀집+희소 하이브리드 검색, 재순위화, 평가 기반 반복까지 프로덕션 RAG 파이프라인을 설계하고 견고하게 다듬어, 그냥 돌아가는 게 아니라 올바른 컨텍스트를 실제로 검색해 오게 만듭니다. 특정 벡터 스토어나 프레임워크에 종속되지 않으며, 방법론은 어떤 임베딩·오케스트레이션 스택에도 이식됩니다.
← Marketplace01 02 03
RAG Pipeline Engineer
by Agentlas
Agent explainer
What this agent actually does
Tool use
If a run needs a plugin or external API, it asks for access first and uses it only within the approved scope.
Result
It produces a result you can review before you apply it.
Best for
What it's good for
RAG 검색 정확도가 낮은데 청킹 문제인지 임베딩 문제인지 평가로 진단해줘.
새 문서 QA 챗봇의 검색 파이프라인을 설계해줘 — 청킹 전략이랑 인덱스 설정까지.
하이브리드 검색 가중치를 어떻게 튜닝해야 하는지 ablation으로 알려줘.
What's inside
What's in this agent
1 skill1 agent1 command
Prerequisites
Before you start
a corpus sample (document types, average length, languages, domain vocabulary)
the expected query distribution (what questions users will ask)
an existing golden eval set, or agreement to build a small one before optimizing
Safety
What it can touch
Access
Files: scoped
Network: none
External API: yes
ONTOLOGY CHIPS
Operational experience and taste compatible with this agent
Hiring the agent and selecting an experience chip are separate decisions. Only verified exact-release matches appear, and none is purchased or attached automatically.
The chip registry could not be loaded. The base agent remains available.
Safety
Inspect everything before it runs
A security scan runs before publish or install, and Agentlas never hosts or proxies models — it runs on your own account and keys.