RAG-10 · Multimodal, table and figure extraction¶
Tables, charts and scanned pages are flattened into broken text, so answers on financial, technical and clinical documents are wrong or missing. Teams commit to these corpora before testing extraction quality and cost on real samples.
- Who reads it: GenAI engineers and RAG leads working on document-heavy corpora, and DevOps engineers running enrichment pipelines
- Outcome: Extraction quality and cost proven per corpus before commitment
- Pack owner: Yogeshwaran (catalog owner role: RAG Lead)
Master catalog entry¶
- Catalog name: Multimodal, Table & Figure Extraction Kit (P3)
- Why it matters: Unlocks technical, financial and clinical corpora competitors cannot handle
- What it removes: Removes the hardest custom work in document-heavy engagements
- Stack: Document Intelligence, GPT-4o vision, Python
- Maturity target: Level 1 — emerging; validate per corpus before committing
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-RAG-10-01 | RAG-10-S1 | Choose the extraction path for tables, figures and scans | Decision guide | Medium | 3 | draft | Kandiah Chockalingam | 2026-12-04 |
| PB-RAG-10-02 | RAG-10-S2 | Extract tables and figures into retrievable chunks | How-to | High | 3 | draft | Kranthi | 2026-12-08 |
| PB-RAG-10-03 | RAG-10-S3 | Per-corpus extraction validation checklist | Checklist | Low | 3 | draft | Kandiah Chockalingam | 2026-11-25 |
| PB-RAG-10-04 | RAG-10-S4 | Handle extraction failures, timeouts and cost overruns in the enrichment pipeline | Runbook | Medium | 4 | draft | Kandiah Chockalingam | 2026-11-17 |