@avisre 100% this.
I learned it the hard way debugging a RAG pipeline, retrieval looked fine on the surface, but traces showed the reranker was silently dropping the right chunks.
No evals = flying blind. What stack are you using for observability ?
I'm Shivam - building AI/LLM systems as a Junior Dev 🧠
What I'm shipping:
→ Agentic workflows with LangChain & LangGraph
→ RAG pipelines with Qdrant
→ LLM-based document intelligence
What's one thing you wish you knew earlier about
building production LLM systems? 👇