Marcos F. Lobo 🇺🇦💙💛 @[email protected]

12.6K posts

Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io banner
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io

Marcos F. Lobo 🇺🇦💙💛 @[email protected]

@arrayexception

Senior Software Engineer & Tech Lead, former @cern. 📰 Writing The Optimist Engineer https://t.co/uDVH8Pw3jd

World 参加日 Haziran 2009
590 フォロー中412 フォロワー
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io
Practical idea: define 5 representative prompts, run them with 2 models and 2 data domains. Measure accuracy, format correctness, and tokens. If just one combination performs clearly better, you’ve already saved time and money.
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io tweet media
English
1
0
0
7
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io
Useful tools to kick off your POC: Promptfoo for CI testing, LangSmith for traceability, and OpenAI Evals for programmable cases. Start with one tool and measure. Then decide if building something internal makes sense.
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io tweet media
English
1
0
1
93
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io
If your org needs governance or traceability, manual prompt testing won’t cut it. Integrate prompt tests into CI: fail the pipeline if a prompt breaks coverage or format. You’ll avoid regressions and production surprises.
Marcos F. Lobo 🇺🇦💙💛 @marcosflobo@hachyderm.io tweet media
English
1
0
0
27