Rough VRAM + index-size sketch for embedding + LLM. Not a full retrieval-pipeline planner.
Limited data: 4 document-count buckets, 4 chunks per doc, no chunking or retriever designer. Order-of-magnitude RAM sketch — not a pipeline planner.
In plain English: sketch a retrieval pipeline (embeddings + LLM) and check whether your hardware can run it.
Check GPU compatibility for your embedding model and LLM, then find the right hardware.