Play 38 — Document Understanding V2 📑🔍
Advanced document processing with layout-aware extraction, cross-doc comparison, and auto-classification.
Beyond Play 06’s basic OCR: auto-classify documents by type, extract fields while preserving layout structure (sections, tables, columns), compare document versions to find changes, and route through automated workflows. Document intelligence meets AI reasoning.
Quick Start
cd solution-plays/38-document-understanding-v2
az deployment group create -g $RG -f infra/main.bicep -p infra/parameters.json
code . # Use @builder for extraction/classification, @reviewer for accuracy audit, @tuner for model routingHow It Differs from Play 06 and Play 15
| Aspect | Play 06 (Doc Intel) | Play 15 (Multi-Modal) | Play 38 (Understanding V2) |
|---|---|---|---|
| Classification | Manual model | Page-type routing | Auto-classify any doc |
| Comparison | None | None | Cross-doc diff + risk |
| Layout | Basic fields | Page-level visual | Section/table-aware |
| Workflow | Extract only | Extract only | Classify → extract → route → archive |
Architecture
| Service | Purpose |
|---|---|
| Document Intelligence | Layout + prebuilt extraction |
| Azure OpenAI (gpt-4o) | Classification, comparison analysis |
| Cosmos DB | Classification results, comparison logs |
| Azure Functions | Workflow orchestration |
Key Metrics
- Classification: ≥92% · Layout F1: ≥88% · Comparison: ≥85% · PII recall: ≥99%
DevKit (Advanced Document-Focused)
| Primitive | What It Does |
|---|---|
| 3 agents | Builder (layout/comparison/classification), Reviewer (accuracy/PII/consistency), Tuner (models/thresholds/cost) |
| 3 skills | Deploy (106 lines), Evaluate (107 lines), Tune (103 lines) |
| 4 prompts | /deploy (understanding pipeline), /test (extraction/workflow), /review (PII/accuracy), /evaluate (classification) |
Cost
| Service | Dev | Prod | Enterprise |
|---|---|---|---|
| Document Intelligence | $0 (Free) | $150 (Standard S0) | $600 (Standard S0) |
| Azure OpenAI | $50 (PAYG) | $350 (PAYG) | $1,100 (PTU) |
| Cosmos DB | $5 (Serverless) | $80 (1000 RU/s) | $400 (5000 RU/s) |
| Azure Functions | $0 (Consumption) | $120 (Premium EP1) | $240 (Premium EP2) |
| Blob Storage | $3 (Hot LRS) | $30 (Hot LRS) | $100 (Hot GRS+WORM) |
| Azure AI Search | $0 (Free) | $250 (Standard S1) | $500 (Standard S2) |
| Key Vault | $1 (Standard) | $5 (Standard) | $15 (Premium HSM) |
| Application Insights | $0 (Free) | $25 (Pay-per-GB) | $100 (Pay-per-GB) |
| Total | $59/mo | $1,010/mo | $3,055/mo |
📖 Full docs · 🌐 frootai.dev/solution-plays/38-document-understanding-v2
FAI Manifest
| Field | Value |
|---|---|
| Play | 38-document-understanding-v2 |
| Version | 1.0.0 |
| Knowledge | F1-GenAI-Foundations, R2-RAG-Architecture, T3-Production-Patterns |
| WAF Pillars | security, reliability, responsible-ai, performance-efficiency |
| Groundedness | ≥ 85% |
| Safety | 0 violations max |
Last updated on