What I learned building an opinionated and minimal coding agent
Lessons I learned while building my own coding agent from scratch.
Summary
Building production RAG systems at scale required custom query generation, chunking strategies, chunk reranking, and injecting metadata to improve results.
Off-the-shelf pipelines produced poor results on large datasets, making iterative system rewrites necessary. Key stack components included vector databases (Azure, Pinecone, Turbopuffer), custom chunking, advanced rerankers, and selective query routing for unsupported questions.
