RAG Trend: Two-Pass Document Processing Balances Cost and Accuracy
llama_index · x · 2026-08-24
Jerry Liu from LlamaIndex outlines a trending two-pass document processing workflow for agent harnesses like Codex and Cowork:
- Fast First Pass: Uses free/OSS lightweight parsing tools to process massive datasets (10-1,000+ files), enabling agents to perform retrieval (grep/semantic) to find relevant context subsets.
- JIT VLM Pass: Applies a VLM-based pass (screenshot + VLM or code) only on the specific pages identified as relevant.
This addresses the speed and cost issues of running VLM-based OCR on massive ad-hoc customer file dumps. It allows for cheap filtering while preserving accuracy for the specific context needed. The post also links to LiteParse, a standalone open-source PDF parser focused on speed.
More from coding & agent
- Gemini CLI Fix: Prevents Output Inflation on Negative maxChars — Kanika0306 · 2026-08-24
- Day 4 of Cloud Agents Migration: Tackling Constant PR Rebasing — jarrodwatts · 2026-08-24
- Graph Engineering organizes multi-agent systems via dynamic structures — Yuyuan Feng · 2026-08-24
- RecVerse agent simulates realistic e-commerce shopping sessions — Jiakai Tang · 2026-08-24
- GitHub Copilot workflows streamline .NET app modernization — tristanbob · 2026-08-24
- SemiAnalysis Open Sources $3M AgentX Benchmark for Agentic Coding Workloads — AccBalanced · 2026-08-24