I keep seeing the same advice in tutorials and on here: "for small projects, just use LlamaIndex." But I'm struggling to understand *why* it's the default recommendation over just using the OpenAI API directly or a more lightweight SDK.
My context: I'm building a simple internal tool that needs to ingest a few hundred PDFs and let users ask questions. I started with the raw OpenAI Assistants API, and it felt... clunky for the document handling. Then I looked at LlamaIndex, and immediately felt overwhelmed by the concepts of nodes, indexes, and query engines. It seems like a huge abstraction layer for what feels like a simple RAG pipeline.
What am I missing that makes it the go-to for small projects? Is it the built-in chunking strategies, the connector ecosystem for PDFs and Notion, or something about the query interface that saves more time than it costs to learn? A concrete example of where it saved you a bunch of boilerplate code would be incredibly helpful.
Right now, the learning curve feels steep for "small," and I'm wondering if I should just push through or if there's a simpler path I'm overlooking.
Connecting the dots.