I do think this would be an idea for Microsoft to improve the SQL Server Management studio, but it could take a while till we see that. Thinkwise could use this for the in-GUI SQL editor I think.

---

Not only for the in-GUI SQL editor. With this question to SQL generation, you could, like Freddy mentions, let the user ask the application questions, in natural language, and the AI would turn that into a query and retreive an answer.



I think for ad-hoc questions from users, this would be an amazing tool! Especially combined with the structured way applications are set up using the Software Factory, you could use the SF as a meta-database, so a user does not have to be aware of the names of tables or fields. The AI knows the translations as well and can use those.



A question like 'How many projects have we running for the customers in Belgium?' should be an easy question to answer. The possibility for end users, is in my opinion endless and very powerful.



I'd strongly encourage that we look into this. @Marcel Zuur, do you have any ideas about this?

---

Leveraging application models and corresponding translation sets in order to process questions and commands in natural language is something we have already looked into over the last two or three years. This concept was actually the basis for an experiment we did with a user interface that used text and voice commands to perform actions (i.e. generate requests to Indicium).



I expect that we will invest more research into this in the future.

---

Leveraging application models and corresponding translation sets in order to process questions and commands in natural language is something we have already looked into over the last two or three years. This concept was actually the basis for an experiment we did with a user interface that used text and voice commands to perform actions (i.e. generate requests to Indicium).



I expect that we will invest more research into this in the future.



What did come out of the experiment, was it successful, and to what degree? Would be really cool to see/know more about this.

---

The proof of concept was very successful. Natural-language understanding is a very good match with model-driven development, because all information needed to train the required machine learning models and to process the results is already available in the Thinkwise model.

You can expect more on this topic in 2020!

---

 

You can expect more on this topic in 2020!

This thread started before the widespread use of LLM. 4 years have passed and this thread would be interesting to get back to the attention. 

What have been the advances in this respect? 

---

I've actually built a little test for this with Python. I connect to a SQL Server database, and leverage a LLM to translate natural language queries into T-SQL. The app uses LangChain to handle the prompt engineering and query generation. For database connectivity, I use SQLDatabase from LangChain's community utilities, which allows me to establish a connection using a SQLAlchemy connection string. The app allows users to input their database connection details, then ask questions about the database in plain English.

It generates and executes the appropriate SQL query, interprets the results, and provides a conversational response as you can see here:
 

The generated queries are rather simple, but with Claude Sonnet 3.5 I've managed to let it generate some more complex T-SQL code.

I think this is deinitely doable within the SF

---

Hey all,

Whilst this is of course super cool to utilize, direct database connection is not possible in a 3-tier architecture, and we went and research a way to have an AI assistant that can talk with Indicium basically. 

What then is possible, is that you can only obtain data that you have rights to based on your role set. For development, you probably have all rights, but for end users this is not the case. 

You will see more on this topic at the Summit 2024

---

I believe that indeed a lot of this Idea will be covered with our plans for UTA as demo’ed during the Summit in October: UTA

---