A modular AI pipeline that converts natural language questions into accurate ERPNext (Frappe) queries, executes them, and returns results in human-friendly language. Built to simplify ERP access for business users with no technical knowledge.
This system allows users to interact with ERPNext using natural questions such as:
“Which suppliers from India are transport providers?”
The pipeline :
- Understands the user intent
- Predicts the relevant ERP doctype and fields
- Generates and executes a valid Frappe query
- Converts the response into a clear, user-friendly sentence using an LLM
Response:
The following suppliers from India provide transport services: IndoTrans Logistics, QuickMove Carriers.
graph TD
A[User Question] --> B[Doctype Detection: RoBERTa]
B --> C[Top Field Detection: SBERT]
C --> D[Query Generation: FLAN-S3]
D --> E[Frappe Query Execution]
E --> F[LLM Natural Language Output]
Contributions are welcome!
You can help by:
- Improving datasets or adding new question styles
- Fixing issues or bugs in the pipeline
- Enhancing query generation or accuracy
To contribute, please open an issue or submit a pull request.
This project is licensed under the MIT License.
#mit