Python
RAG Index
RAG lab: PDF to nodes, SummaryIndex and VectorStoreIndex, router between summary and detail. LLM and embeddings via NVIDIA NIM. Index on disk.
View code ↗What it solves
I wanted a RAG pipeline without LangChain on top. An orchestrator loads PDFs, splits to nodes, builds summary and vector indexes, and RouterQueryEngine picks. Agent mode with tool calling is experimental; not every NVIDIA model supports it. Single-doc or one index per file.
Stack
- Python
- LlamaIndex
- NVIDIA NIM
- RAG
Infra
Python 3.10+, LlamaIndex, NVIDIA NIM. Local persist in storage/. Key in .env. No deploy; the link is the repo.