PPIDeclarePilotSandbox

Pipeline to:

• mine DECLARE constraints from event logs,
• generate KPI proposals with an LLM,
• generate PPINot code,
• execute PPIs on the log,
• produce reports and experiment artifacts.

Project Structure

• src/declareppipilot/: core library (pipeline.py, models.py, llm.py, prompts/schemas)
• tests/: runnable integration scripts for the pipeline
• experiments/: RQ experiment scripts (rq1 ... rq5)
• data/logs/: input logs (e.g. DomesticDeclarations.xes)
• outputs/: generated artifacts
• requirements.txt: Python dependencies

Requirements

pip install -r requirements.txt

Depending on provider, configure credentials:

• OpenAI: OPENAI_API_KEY
• Anthropic: ANTHROPIC_API_KEY
• LM Studio: LMSTUDIO_BASE_URL and LMSTUDIO_MODEL (optional key env)

Quick Start

Run from project root.

1. Run full pipeline test

python tests/test_full_pipeline.py

The script auto-detects available providers and writes outputs into outputs/.

2. Run step-by-step pipeline test

python tests/test_declareppipilot.py

3. Run web dashboard

uvicorn webapp.main:app --reload

Then open:

• http://127.0.0.1:8000/

Dashboard features:

• upload .xes log
• set provider/model (and optional API key)
• optional time grouping via quantity + unit (internally mapped to xWE, xME, xYE)
• optional min_support
• KPI dashboard grouped by business goal with AI-generated readable PPI text and technical details on demand

Using the Library

If running scripts outside this repo root, set PYTHONPATH:

export PYTHONPATH=src

Main API entry points:

• run_full_pipeline(...)
• mine_declare_model_and_constraints_df(...)
• generate_kpis_from_declare_table(...)
• generate_ppinot_from_kpi_set(...)
• execute_ppinot_kpis(...)
• build_kpi_execution_report(...)

Experiments

RQ scripts and usage are documented in:

• experiments/README.md

They support multiple providers (openai, anthropic, lmstudio) and save timestamped outputs under outputs/experiments/....

Notes

• run_full_pipeline now includes automatic generation of human-readable PPI definitions from kpi_metric_str.
• Generated code execution includes safety checks before exec.