LlamaIndex pitches two-pass just-in-time OCR: parse data rooms in 32s, zoom only where needed
llama_index · x · 2026-09-15
LlamaIndex's official account argues that parsing every page upfront is slow and expensive for agents working through ad-hoc data rooms, since most pages never get read. The proposed pattern is two-pass just-in-time OCR:
- Pass 1: LiteParse (free, OSS, Rust, 50+ formats) does a fast layout-aware parse — spatial text, bounding boxes, headings, tables, and a per-page complexity flag — processing a full data room in 32 seconds.
- Pass 2: LlamaParse zooms in only on flagged pages, returning cell-level tables, bounding boxes, and confidence scores, while the rest stays as background.
The post notes pypdf and pdftotext can't handle the first pass well, and full upfront parsing can't be cheap — the two-pass approach gets you both speed and precision.
More from coding & agent
- Claude Mods: Anthropic to ship function hooks-based plugins for Claude Code in weeks — bcherny · 2026-09-15
- The hidden cost of failed agent runs: one task may burn twice the credits on Codex — ToneAromatic178 · 2026-09-15
- OpenAI Agents API enters public beta: managed cloud agents on the Codex harness — craigsdennis · 2026-09-15
- Interview with the Claude Code team on building it while models keep outpacing engineering — EricBuess · 2026-09-15
- Before buying a faster model, check your agent's traces: serial API calls eat latency — gethackteam · 2026-09-15
- ElevenLabs MCP adds voice, music, image, and video generation — lukeharries · 2026-09-15