Flowset Demo — Loan Scoring with Operaton & AI Fraud Check

A Spring Boot application that demonstrates end-to-end BPMN process automation using Operaton 7.

The process models a loan application review pipeline: it calculates a credit score, evaluates it against a DMN decision table, runs a parallel AI-powered fraud check via OpenAI, and routes the result to a human reviewer in Operaton Tasklist.

---

Tech Stack

Layer	Technology
Language	Java 17
Framework	Spring Boot 3.4.4
BPM Engine	Operaton Platform 1.1.1
Decision Engine	Operaton DMN
Database	H2 (file-based, auto-created)
AI	OpenAI API (gpt-4o-mini)
Build	Gradle 8 (wrapper included)

---

Prerequisites

• Java 17+ installed and on PATH
• An OpenAI API key (required for the AI fraud-check step)

No other installations needed — Operaton runs embedded inside the Spring Boot app, and H2 is created automatically on first run.

---

Quick Start

1. Clone the repository

git clone https://github.com/your-org/flowset-operaton-demo.git
cd flowset-operaton-demo

2. Set your OpenAI API key

Open src/main/resources/application.properties and replace the placeholder:

ai.openai.api-key=<your-key-here>

If you skip this step the application will still start, but the AI Fraud Check step will fail and create a Operaton incident.

3. Run the application

./gradlew bootRun

The application starts on http://localhost:8080.

---

Operaton UI

URL	Purpose
http://localhost:8080/operaton/app/tasklist	Start and review loan applications
http://localhost:8080/operaton/app/cockpit	Monitor active process instances
http://localhost:8080/operaton/app/admin	Manage users and authorizations

Login: admin / Password: admin

---

How to Run the Process

1. Open Tasklist → click Start process → choose loan-scoring-v1
2. A start form appears — select one of the pre-loaded applicants from the dropdown (see table below) → click Submit
3. The process runs automatically:
   • loads applicant data
   • calculates a credit score
   • in parallel: evaluates the DMN rule + calls OpenAI for fraud analysis
4. Once both parallel branches complete, a Review Application task appears in your Tasklist
5. Open the task — you see all applicant data, the credit score, the rule decision, and the AI recommendation
6. Choose Approve or Reject → Submit
7. The process completes (approval or rejection is logged to the console)

---

Demo Applicants

Six applicants are pre-loaded in ApplicantRepository. Each is designed to produce a different outcome:

Applicant ID	Name	Age	Income	Loan	Credit History	Expected outcome
app_low_risk_01	Emma Thompson	36	$210,000	$140,000	GOOD	High score → Approve
app_premium_01	Olivia Chen	48	$260,000	$150,000	EXCELLENT	Max score → Approve
app_borderline_01	Sofia Martinez	33	$110,000	$90,000	GOOD	Mid score → borderline
app_no_history_01	Liam Johnson	24	$68,000	$230,000	NONE	Young + no history → low score
app_senior_large_loan_01	Michael Brown	62	$180,000	$400,000	GOOD	Large loan → low score
app_high_risk_01	Carlos Vega	27	$42,000	$780,000	DELINQUENT	Low score + delinquent → Reject

---

Process Architecture

Start Event (form)
    │
    ▼
Load Applicant Data       ← reads applicant from in-memory repository
    │
    ▼
Calculate Score           ← scoring formula based on income/loan ratio, age, credit history
    │
    ▼
Parallel Gateway ─────────────────────────────────┐
    │                                             │
    ▼                                             ▼
Evaluate Rule Decision                    AI Fraud Check
(DMN: loan-decision.dmn)                  (OpenAI gpt-4o-mini)
ruleDecision = Approve / Reject           aiRecommendation + aiRiskLevel
    │                                             │
    └──────────────────┬───────────────────────────┘
                       │
                       ▼
             Review Application       ← human task in Tasklist
             (review-form-v1.form)
                       │
                       ▼
             Exclusive Gateway
             ┌─────────┴─────────┐
             ▼                   ▼
       Notify Approval     Notify Rejection
       (log)               (log)
             │                   │
             ▼                   ▼
          End Event           End Event

Scoring Formula

score = (income / loanAmount) × 50
      + ageFactor   (age 25–60 → +10,  age < 25 → +0,  age > 60 → +5)
      − penalty     (EXCELLENT → 0,  GOOD → −5,  NONE → −10,  DELINQUENT → −20)

Clamped to [0, 100].  Score ≥ 70  →  DMN outputs "Approve".

DMN Decision Table (loan-decision.dmn)

Credit Score	Rule Decision
≥ 70	Approve
< 70	Reject

---

Project Structure

src/main/java/io/flowset/demo/
├── FlowsetDemoApplication.java          # Spring Boot entry point
├── delegate/
│   ├── LoadApplicantDataDelegate.java   # Loads applicant, sets process variables
│   ├── ScoringDelegate.java             # Computes credit score
│   ├── AIFraudCheckDelegate.java        # Calls OpenAI, writes recommendation + risk level
│   ├── ApprovalNotifyDelegate.java      # Logs approval
│   └── RefusalNotifyDelegate.java       # Logs rejection
├── model/
│   └── Applicant.java                   # Immutable record
├── repository/
│   └── ApplicantRepository.java         # In-memory store with 6 demo applicants
├── security/
│   └── WebSecurityConfiguration.java    # Security + CORS (open for demo)
└── variable/
    └── VariableConstants.java           # All process variable name constants

src/main/resources/
├── application.properties               # App + Operaton + OpenAI config
├── processes/
│   ├── loan-scoring-v1.bpmn             # Main loan scoring process
│   └── simple-process.bpmn             # Minimal Hello World process (for reference)
├── decisions/
│   └── loan-decision.dmn               # DMN rule: score → Approve / Reject
└── process-forms/
    ├── start-form-v1.form              # Applicant selector (start event form)
    └── review-form-v1.form             # Human review form with all computed variables

---

Configuration Reference

All configuration lives in src/main/resources/application.properties:

# H2 file-based database (auto-created in ./h2/)
spring.datasource.url=jdbc:h2:file:./h2/operaton-h2-database

# Operaton admin credentials
operaton.bpm.admin-user.id=admin
operaton.bpm.admin-user.password=admin

# OpenAI integration
ai.openai.api-url=https://api.openai.com/v1/chat/completions
ai.openai.model=gpt-4o-mini
ai.openai.timeout-seconds=15
ai.openai.api-key=<your-key-here>

The H2 database file is created automatically at ./h2/operaton-h2-database.mv.db and is excluded from git.