Traderiser

B2B Platform

Overview

To develop a platform that converts natural language to SQL queries, the client wants a platform where we can add databases as data sources and for retrieving the data instead of using complex queries in SQL we can simply ask questions in plain English, and the platform will provide all answers needed. The problem is widely popular in the Natural Language Processing domain as a “Text-to-Code” conversion. We Created the Hybrid approach to tackle the problem where we are using statistical as well as deep learning to solve it.

Natural language processing to solve the Text-to-code problem
Deep learning-based as well as statistical models aka Hybrid approach
Google Bigquery for warehousing the data sources
Django APIs for easy interaction with the platform
Google IAM authentication in GCP

Technology

Google Cloud Platform
Deep Learning
Django
Ln2SQL

Key Technical Challenges:

Creation of a model that works with cross-domain data sources with different domain knowledge
Warehouse in Google Bigquery in a way that works with PostgreSQL, MySQL and MsSql database servers
Accuracy in predicting query is the most crucial part of the Hybrid approach and provide feedback to the model.

Business + Technical Points:

Around 70% accuracy with Provided data sources by the client.
A scalable model that improves over time as more and more queries in the future will come with different data sources.
The central warehouse that works with a different type of SQL based servers.

Our Work