Play 06 — Document Intelligence 📄

Extract, classify, and structure document data with OCR + LLM.

Feed PDFs, invoices, receipts, and forms into Azure Document Intelligence for OCR, then GPT-4o extracts structured fields into typed JSON. Handles multi-page documents, handwriting, tables, and stamps.

Quick Start


cd solution-plays/06-document-intelligence
az deployment group create -g $RG -f infra/main.bicep -p infra/parameters.json
code .  # Use @builder for extraction, @reviewer for accuracy audit, @tuner for throughput

Key Metrics

Field extraction: ≥95% (prebuilt), ≥90% (custom) · Processing: <10s/page · PII recall: ≥99%

DevKit

Primitive	What It Does
3 agents	Builder (OCR/extraction), Reviewer (accuracy/PII audit), Tuner (model selection/cost)
3 skills	Deploy (117 lines), Evaluate (110 lines), Tune (105 lines)

Architecture

Rendering diagram…

📐 Full architecture details — data flow, security architecture, scaling guide

Cost Estimate

Service	Dev/PoC	Production	Enterprise
Document Intelligence	$0 (Free)	$100 (Standard S0)	$400 (Standard S0)
Azure OpenAI	$40 (PAYG)	$250 (PAYG)	$900 (PTU Reserved)
Blob Storage	$2 (Hot LRS)	$25 (Hot LRS)	$80 (Hot GRS)
Container Apps	$10 (Consumption)	$80 (Dedicated)	$250 (Dedicated HA)
Cosmos DB	$5 (Serverless)	$50 (Autoscale)	$200 (Autoscale)
Key Vault	$1 (Standard)	$3 (Standard)	$10 (Premium HSM)
Application Insights	$0 (Free)	$25 (Pay-per-GB)	$80 (Pay-per-GB)
Event Grid	$0 (Free)	$2 (Basic)	$10 (Basic)
Total	$58/mo	$535/mo	$1,930/mo

💰 Full cost breakdown — per-service SKUs, usage assumptions, optimization tips

📖 Full docs · 🌐 frootai.dev/solution-plays/06-document-intelligence

FAI Manifest

Field	Value
Play	`06-document-intelligence`
Version	`1.0.0`
Knowledge	F1-GenAI-Foundations, T3-Production-Patterns
WAF Pillars	security, reliability, responsible-ai
Groundedness	≥ 85%
Safety	0 violations max