An AI-powered tool that analyzes business process descriptions and identifies Lean waste categories with actionable improvement recommendations.
Built with Azure OpenAI and grounded in Lean Six Sigma methodology. Designed as a portfolio project demonstrating how AI can accelerate process improvement work that traditionally requires expert facilitation.
Process improvement initiatives typically require a trained Lean facilitator to observe workflows, conduct value stream mapping sessions, and identify waste manually. This is time-intensive, expensive, and inconsistent across teams.
This tool allows any team to describe their current process in plain language and receive a structured waste analysis instantly — identifying which of the 8 Lean waste categories are present, why they exist, and what to do about them.
- User submits a process description as plain text (e.g., "Our invoice approval process requires the manager to print the invoice, sign it, scan it, and email it to Accounts Payable who re-enters the data manually.")
- Azure OpenAI analyzes the description against the 8 Lean waste categories (DOWNTIME)
- Tool returns a structured waste analysis: which wastes are present, severity, root cause hypothesis, and improvement recommendations
- Output is saved as JSON and a human-readable Markdown report
- Optional: batch mode processes multiple process descriptions from a CSV
| Code | Waste | Description |
|---|---|---|
| D | Defects | Errors requiring rework, correction, or scrap |
| O | Overproduction | Producing more than needed, sooner than needed |
| W | Waiting | Idle time waiting for the next step, approval, or input |
| N | Non-utilized talent | Skills, knowledge, and ideas not being used |
| T | Transportation | Unnecessary movement of information or materials |
| I | Inventory | Excess work-in-progress, backlog, or unused data |
| M | Motion | Unnecessary physical or digital movement by people |
| E | Extra processing | Doing more work than the customer or process requires |
[Process Description Input]
|
| Plain text (single or batch CSV)
v
[Input Validator]
|
v
[Azure OpenAI GPT-4o]
System prompt: Lean waste analysis framework
User message: process description
|
v
[Structured Output Parser]
- Waste categories detected
- Severity rating (High / Medium / Low)
- Root cause hypothesis
- Improvement recommendations
|
+-------------------+
| |
v v
[JSON Output] [Markdown Report]
|
v
[Optional: Power BI / CSV Dashboard]
| Layer | Tooling |
|---|---|
| Cloud platform | Azure |
| LLM analysis | Azure OpenAI (GPT-4o) |
| Structured output | OpenAI JSON mode / response schema |
| Runtime | Python 3.10+ |
| Reporting | Markdown + optional CSV/Power BI |
| Version control | GitHub |
process-waste-detector/
├── README.md
├── LICENSE
├── docs/
│ ├── lean-framework.md # DOWNTIME methodology reference
│ ├── prompt-design.md # System prompt rationale and iterations
│ ├── setup.md # Environment setup guide
│ └── decisions.md # Architecture decision log
├── src/
│ ├── analyzer.py # Core waste analysis logic
│ ├── batch.py # Batch mode: process CSV of descriptions
│ ├── reporter.py # Formats output as Markdown report
│ └── utils/
│ ├── config.py # Env var loader
│ └── logger.py # Structured logging
├── data/
│ ├── samples/ # Sample process descriptions for testing
│ └── outputs/ # Generated reports and JSON results
├── tests/
│ ├── test_analyzer.py
│ └── test_reporter.py
├── .env.example
├── requirements.txt
└── .gitignore
git clone https://github.com/Automater89/process-waste-detector.git
cd process-waste-detector
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
cp .env.example .env
# Add your Azure OpenAI credentials to .env
# Analyze a single process
python src/analyzer.py --input "Our invoice approval requires printing, manual signature, scanning, and re-entry into the ERP system."
# Analyze a file
python src/analyzer.py --file data/samples/sample_process.txt
# Batch mode
python src/batch.py --csv data/samples/processes.csvInput:
Our invoice approval process requires the manager to print the invoice, sign it, scan it, and email it to Accounts Payable who re-enters the data manually into the ERP.
Output:
{
"process_description": "Our invoice approval process...",
"wastes_detected": [
{
"category": "Extra Processing",
"code": "E",
"severity": "High",
"evidence": "Manual re-entry of data already present in the invoice",
"root_cause": "Lack of system integration between approval workflow and ERP",
"recommendation": "Implement e-signature with direct ERP integration via Power Automate or AP automation platform"
},
{
"category": "Motion",
"code": "M",
"severity": "Medium",
"evidence": "Print, sign, scan cycle adds physical steps with no value",
"root_cause": "Paper-based approval process not updated for digital workflow",
"recommendation": "Replace physical signature with digital approval in existing document management system"
},
{
"category": "Waiting",
"code": "W",
"severity": "Medium",
"evidence": "Invoice sits idle between manager signature and AP receipt",
"root_cause": "Email handoff with no SLA or routing automation",
"recommendation": "Automate routing with Power Automate; set approval SLA with escalation triggers"
}
],
"overall_severity": "High",
"priority_action": "Eliminate manual re-entry by integrating e-signature with ERP",
"estimated_waste_score": 7.2
}- Configure Azure OpenAI credentials
- Validate connectivity with config.py
- Confirm GPT-4o deployment responds to test prompt
- Build system prompt with full DOWNTIME framework
- Implement JSON-mode structured output
- Test against 5 sample process descriptions
- Format JSON output as readable Markdown report
- Save to
data/outputs/with timestamp - Include severity summary and priority action at top
- Accept CSV with multiple process descriptions
- Run analyzer on each row
- Output combined report and individual JSON files
- Add 10 sample processes across industries (HR, finance, manufacturing, IT)
- Record Loom walkthrough demo
- Add LinkedIn project summary
- Link to Agent Showcase: https://automater89.github.io/Agent-Showcase/
- HR onboarding process audit
- IT service desk ticket handling review
- Finance invoice and expense approval workflows
- Manufacturing quality inspection steps
- Any team wanting a fast Lean assessment without a consultant
This project is grounded in certified Lean Six Sigma Black Belt (ICBB) methodology:
- DOWNTIME waste framework applied across HR, benefits, and IT operations
- Reduced processing time by 60% through benefits automation in a public sector HR environment
- Implemented audit automation reducing bi-weekly file validation from manual to automatic
- Eliminated manual consolidation in analytics pipelines via SharePoint-Forms-Excel integration
The prompts, severity rubric, and recommendation patterns reflect real process improvement work — not generic templates.
- benefits-faq-agent — RAG-based HR benefits Q&A agent
- azure-doc-agent — Document extraction and agent workflow pipeline
- Agent Showcase — Live portfolio of AI and automation projects
Current phase: scaffold and planning.
MIT