We love metrics. But as developers, we often struggle to measure abstract concepts like code tech-debt, code review harshness, or documentation clarity.
Ana Fikir: (Burada LLM'e bir metin verip "Bunu -1 (çok negatif) ile 1 (çok pozitif) arasında puanla ve sadece JSON dön" dediğin basit bir prompt senaryosundan bahset.)
Tartışma Sorusu: If we can turn every qualitative data into a quantitative graph, where does human intuition fit in? Let's discuss in the comments!
I have already deployed this system, but I'm curious about your edge cases:
How would you handle a situation where the human emotion is deeply sarcastic? Sarcasm completely breaks standard sentiment metrics. How should the agent detect the 'subtext' without getting confused?
https://github.com/bestekarx/humetric
Let's discuss in the comments! 👇
Top comments (0)