A beginner-friendly CrewAI + Groq + Streamlit prototype for construction payment review.
The agents analyze evidence, calculate, reconcile and flag exceptions. They do not release money, call an ERP payment API, approve a payment, or make the final decision.
- Intake & Evidence
- Contract Compliance
- Measurement & Valuation
- Payment Reconciliation
- Risk & Exceptions
- Human Review Brief
Each agent is in its own Python file.
The agents use CrewAI tools for:
- Reading uploaded PDF/DOCX/XLSX/CSV/TXT/MD documents
- Payment arithmetic
- Required-document checks
- Audit-event recording
This repository is designed for Streamlit Community Cloud.
Set this secret:
GROQ_API_KEY = "your_groq_api_key"Main entrypoint:
app.py
Python version:
3.12
The project pins the core package versions in requirements.txt.
The model is configured through CrewAI's Groq/LiteLLM provider route using groq/openai/gpt-oss-120b; the standalone groq package is included for compatibility with the Groq SDK ecosystem.