Ravi Theja

Background

Wrote much of LlamaIndex’s empirical RAG-evaluation series — the work that put numbers behind retrieval-tuning advice rather than intuition.

Articles in This Vault

• NVIDIA Research - RAG with Long Context LLMs — retrieval helps long-context models too; 5–10 chunks optimal

Key Contributions

• Popularised measuring RAG retrieval with hit rate and MRR via LlamaIndex’s evaluation modules
• Documented the “lost in the middle” retrieval-budget limit — see Long-Context RAG

Concepts

• Long-Context RAG · RAG · Reranking · MRR

Tools

• LlamaIndex