Is AI smart enough to deliver data insights if we ask questions in plain English? AI2SQL claims to be a part of the answer.
So lets try it out!
About AI2SQL
AI2SQL can generate SQL statements out of data queries in plain English.
Under the hood it uses OpenAI libraries (GPT-3 and Codex).
OpenAI is an AI research and deployment venture. Their mission is to ensure that artificial general intelligence benefits all of humanity. OpenAI’s other project is DALl.E2. DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language. Besides they have other exciting NLP functionality.
We’ll try the other functionalities some other day. Let’s get back to AI2SQL.
Use Case
WBL Bank offers credit cards and mortgages. The product managers ask a lot of questions; and the data analysts fetch the answers. Both are quite fed up of each other ! The data analysts want the product managers to self-service; the product managers say they have no tools to access the information themselves.
Let solve this using AI2SQL.
Data Structure
The back end database has the following tables:
customer_master
credit_card_master
credit_card_txns (transactions)
mortgage_master
mortgage_txns (transactions)
The ER diagram of the tables is below.
Setting up AI2SQL
The set-up is quite simple. Need to specify the table name, table description and the column names.
Let’s Ask Some Questions
Asking a question is easy. You just ask the question and click on “Generate SQL”.
I tried 5 questions.
Verdict
AI2SQL (or rather the underlying GPT-3 and Codex libraries) are excellent. As long as the column names are explanatory, the SQLs generated are accurate.
So what’s next
To use this as an end to end solution for WBL, we need the following additional components:
Connection to a back end DB
Actually running the query and fetching results
Juicy topics for a future post !!