Step-by-Step Teaching Guide

Step 1: Understanding the Problem (The "Why")

The Core Challenge

Enterprises receive millions of documents daily — invoices, contracts, reports, etc.

These documents are:

• 📄 Messy and unstructured (PDFs, scanned files, images)
• 💼 Packed with critical business data (financial info, customer details, legal terms)
• ⏳ Requiring manual reading before the data can be used

The Bottleneck Visualized

Document Arrives → Someone Reads It → Extracts Data → Business Decision
                        ↑
                   THIS IS SLOW
                   (and gets worse as volume grows)

Key Insight: The manual extraction step is the bottleneck. As document volume scales, costs and delays explode.

---

Step 2: The Solution — Document Intelligence REST APIs

What Are They?

Databricks released real-time REST APIs that automatically:

• Parse documents
• Classify documents
• Extract structured data from documents

Why REST APIs Specifically?

REST APIs allow developers to:

• Call document processing from any application
• Integrate into existing workflows
• Process documents in real-time (not batch)
• Power both human-facing apps and AI agents

---

Step 3: The Three Core APIs (+ 1 Beta)

Think of these as an assembly line for document understanding:

🔧 API #1: ai_parse_document

---

🏷️ API #2: ai_classify

---

🔍 API #3: ai_extract

---

🔀 API #4: Decisions API (Beta)

---

Step 4: How the APIs Work Together — A Real Example

Use Case: Contract Intake for a Law Firm

STEP 1: Lawyer uploads a contract
            ↓
STEP 2: ai_parse_document
        → Converts PDF into structured content
            ↓
STEP 3: ai_classify
        → "This is an NDA" (not an MSA or SOW)
        → App knows which review form to display
            ↓
STEP 4: ai_extract
        → Pulls: parties, effective date, renewal terms, notice period
        → Provides citations so lawyer can verify each value
            ↓
STEP 5: Review form opens with fields PRE-FILLED
        → Lawyer reviews and approves (human-in-the-loop)

Result: What used to take 30+ minutes of manual reading now happens in real-time

---

Step 5: Two Major Use Cases

Use Case Type 1: 👤 Human-in-the-Loop Applications

• Built with Databricks Apps
• Human reviews AI-extracted data
• Example: Lawyer reviewing pre-filled contract form
• Best for: High-stakes decisions requiring human verification

Use Case Type 2: 🤖 Autonomous AI Agents

• Built with Agent Bricks
• Agent calls APIs to understand documents before making decisions
• No human required in the loop
• Best for: High-volume, routine document processing

Human-in-the-Loop:  Document → APIs → Pre-filled Form → Human Reviews → Decision
Autonomous Agent:   Document → APIs → Agent Analyzes → Agent Decides → Action

---

Step 6: Key Concepts Summary

---

Step 7: Quick Knowledge Check ✅

Test yourself with these questions:

1. What problem do Document Intelligence APIs solve?
2. Which API would you use to determine if a document is an invoice vs. a purchase order?
3. Which API would you use to pull out the total amount from an invoice?
4. What is the difference between a human-in-the-loop app and an autonomous agent workflow?
5. Why are citations an important feature of ai_extract?

---

Summary Diagram

                    DOCUMENT ARRIVES
                          ↓
              ┌─────────────────────┐
              │  ai_parse_document  │ → Structured Content
              └─────────────────────┘
                          ↓
              ┌─────────────────────┐
              │    ai_classify      │ → Document Type
              └─────────────────────┘
                          ↓
              ┌─────────────────────┐
              │    ai_extract       │ → Key Fields + Citations
              └─────────────────────┘
                          ↓
              ┌───────────┴───────────┐
              ↓                       ↓
    Human Reviews App          AI Agent Acts
    (Databricks Apps)          (Agent Bricks)

Bottom Line: Document Intelligence APIs eliminate the manual bottleneck between document arrival and business action — enabling faster, scalable, and more accurate document workflows.