Inspiration
Read about Uber's QueryGPT which is trained on their databses and minimizes errors in query generation. I know that giving your database schema to an llm is not good for privacy for lots of organizations.
What it does
Learns about the database's schema, trains a gnn based on the relationships of the tables. User enters a natural language query and it suggests tables, then the user selects relevant tables then a sql query is generated.
How we built it
Python backend to train the gnn and handle processing on the edge. React frontend
Challenges we ran into
Generated queries are not super accurate always due to the simplicity so far. After some more better integration and training it can be better easier.
Accomplishments that we're proud of
Built a framework for basic database learning and privacy based generation of sql query from natural language for any mysql databse
What we learned
How to train gnn, text-to-sql generation is not super easy to train in a small amount of time.
What's next for GraphSQL Tool
Better learning for the models to increase the accuracy of the queries
Log in or sign up for Devpost to join the conversation.