AI 엔지니어링 루프, 처음부터 끝까지.
이것은 작은 TypeScript 샘플 애플리케이션인 Dad IT Support Agent를 기반으로 한 단계별 Langfuse 워크숍입니다. 이 워크숍은 Langfuse를 사용한 전체 AI 엔지니어링 루프를 다룹니다: 트레이싱, 프롬프트 관리, 모니터링, 데이터셋, 실험, 평가.
What you will do
Have the workshop app running locally with both OpenAI and Langfuse credentials in place. From here you can skim 01-base-app, then start building in 02-tracing.
You can customize your experience by changing the phone specs in support-data.ts file. Adding your dad's phone information means, you will get replies for the right type of phone.
This is the blank slate for the tracing step — same code as checkpoint/01-base-app, with no Langfuse wiring yet. The Langfuse packages are already in package.json — run npm inst...
You have a working traced app. The system prompt lives as a constant called SYSTEMPROMPT in src/server/support-agent.ts and is used directly as the system message.
You have a traced app with optional Langfuse-managed prompts. Every chat turn lands in Langfuse as a nested trace.
You have a traced, attributed, monitored app. data/seed-dataset.json and scripts/seed-dataset.ts are already in the repo at this checkpoint.
Your dataset is seeded in Langfuse. scripts/run-dataset.ts is already in the repo.
Your app is traced, monitored, has a hosted dataset, and at least one experiment run with both keywordoverlap and correctness scores. Now you make a change to the app and rerun ...
You have walked through every loop step.