This project demonstrates an agentic workflow using LangGraph to generate a Business Process Review (BPR) document from:
- A client conversation transcript
- A discovery questionnaire
- A BPR template
The output is validated against a ground truth BPR for feedback and optional refinement.
All these documents have been generated using LLMs and are dummy in nature and are not a copy of a real document.
- Modular agent design (Ingestor, Extractor, Synthesizer, Evaluator, Refiner)
- LangGraph-based stateful workflow with conditional refinement
- LangSmith tracing support for debugging and evaluation
- CLI interface for end-to-end execution
agentic_bpr_poc/ ├── agents/ # Individual agents (LLM-backed)
├── workflow.py # LangGraph DAG definition
├── main.py # Entry point CLI
├── .env.example # LangSmith/OpenAI key management
├── requirements.txt # Dependencies
└── README.md # Docs
- Clone Repo & Install Dependencies
- uv pip install -r requirements.txt