How-toUpdated 2026-06-28 · 17 min read · by RTXsparks Lab

Building a Local RAG Pipeline on RTX Spark

Build a private, high-throughput RAG pipeline on RTX Spark using bge-m3 embeddings, Qdrant, and Llama-3.1-70B. End-to-end with benchmarks.

Stack

Qdrant + bge-m3 + vLLM.

Frequently asked questions

Which reranker?

bge-reranker-v2-m3; adds 40ms but +12% recall@10.

Related guides