Back to thoughts
Article

2025-12-04

1 min read

Notes on RAG Quality

Most RAG problems are retrieval problems, not generation problems.

RAGEvaluationLLMs

Common Mistake

Teams often tune prompts while ignoring weak retrieval corpora.

Better Order

  1. fix chunking strategy
  2. verify document quality
  3. evaluate retrieval precision/recall
  4. then tune generation behavior

Rule of Thumb

If the right context never reaches the model, better prompts will not save the answer.

Related

Building AI Applications in 2025

My thoughts on the current state of AI development and what it means for software engineers. From RAG systems to agent frameworks, here's what I've learned.

Status Terminal

Mathieu Foko

Mathieu Foko

Software Engineer + AI Student

Open to collaborations

Current Focus

  • ▸Building AI applications at Heimat24
  • ▸Studying AI in Germany
  • ▸Shipping Swift + Next.js apps

I build software people actually use. Most days I am shipping AI workflows, retrieval systems, and product features, then heading back to class to study modern AI systems.