Setting Up Calculation Rules in Dust for BigQuery Integration

Youcef Zeghiche · 2025-03-11 10:25:22.091986+00

Hello, I have connected Dust with my Bigquery table and it's so great ! Now I don't know how/where to give rules about my table. Indeed, Dust can't guess which columns choose to make a rate or calcul field. For exemple I would like ask to dust : "What is my conversion rate ? " but Dust dosen't know which columns choose to make the calcul There is a place where I can give all calculated rules which are related to the questions coming from no-data users ? Thanks in advance

Setting Up Calculation Rules in Dust for BigQuery Integration | Dust Community

11 comments

· Sorted by Oldest

Abboud Hammour
·
·
Hi Youcef Zeghiche! I would suggest putting them in the instructions of the agent (i.e. maybe a section like "Business Logic" and then you have the calculations under. If you have a lot, you can also put them in a text file and just give access to the agent via the Search tol
Abboud Hammour
·
·
I'd also recommend using the "Analyst" template for the instructions if using BQ
Youcef Zeghiche
·
·
Great, thanks! Abboud Hammour, I have about ten analytical tables (transformed, cleaned, and ready to be queried) with different levels of granularity and different calculation rules. Do you recommend creating a single text file for all tables or one for each table? Should I also follow a specific format?
Abboud Hammour
·
·
Single text file should be fine! On format, nothing specific but I'd suggest in text (vs. tabular)) and as clear as possible (i.e. if using a word doc, table names as headings, column names as sub headings, clear descriptions, separate section for business logic)
Abboud Hammour
·
·
But otherwise as long as it's clear & you mention in the instructions to use that file to access info on the logic, the agent should be able to use and find it

Youcef Zeghiche

Oh super thanks ! for exemple this format is good for dust ? do you think dust understand my business logic which is displayed below ?

sales_transactions
This table contains successful transactions.

Table Fields:

0.transaction_attempt
Description: Unique identifier for the attempt of the transaction.

1.transaction_id
Description: Unique identifier for the transaction.

2.customer_id
Description: Unique identifier for the customer who made the transaction.

3.product_id
Description: Unique identifier for the purchased product.

4.transaction_date
Description: Date and time of the transaction.

5.quantity
Description: Number of units of the product purchased.

6.total_amount
Description: Total amount of the transaction in local currency.

payment_method
7.Description: Payment method used (card, cash, bank transfer, etc.).

8.store_location
Description: Location of the store where the transaction took place.

Business logic
1. conversion_rate : transaction_attempt / transaction_id

2. Etc....

Abboud Hammour
·
·
Yes I think that would be good (maybe add one line after business logic "this contains business logic for calculations of columns across tables")
Youcef Zeghiche
·
·
Oh thanks ! I have a last question : How Can I help dust to anticipate unpredicted user inputs like "give me transformation rate ?" instead of "conversion rate" ?
Youcef Zeghiche
·
·
I'm sure that i will have many semantic problems
Abboud Hammour
·
·
No perfect science unfortunately other than putting common terms in the instructions, but LLMs are pretty good at guessing even without so I'd test a few!
Youcef Zeghiche
·
·
Okay thank you very much !

Abboud Hammour
·
·
Hi Youcef Zeghiche! I would suggest putting them in the instructions of the agent (i.e. maybe a section like "Business Logic" and then you have the calculations under. If you have a lot, you can also put them in a text file and just give access to the agent via the Search tol
Abboud Hammour
·
·
I'd also recommend using the "Analyst" template for the instructions if using BQ
Youcef Zeghiche
·
·
Great, thanks! Abboud Hammour, I have about ten analytical tables (transformed, cleaned, and ready to be queried) with different levels of granularity and different calculation rules. Do you recommend creating a single text file for all tables or one for each table? Should I also follow a specific format?
Abboud Hammour
·
·
Single text file should be fine! On format, nothing specific but I'd suggest in text (vs. tabular)) and as clear as possible (i.e. if using a word doc, table names as headings, column names as sub headings, clear descriptions, separate section for business logic)
Abboud Hammour
·
·
But otherwise as long as it's clear & you mention in the instructions to use that file to access info on the logic, the agent should be able to use and find it

Youcef Zeghiche

Oh super thanks ! for exemple this format is good for dust ? do you think dust understand my business logic which is displayed below ?

sales_transactions
This table contains successful transactions.

Table Fields:

0.transaction_attempt
Description: Unique identifier for the attempt of the transaction.

1.transaction_id
Description: Unique identifier for the transaction.

2.customer_id
Description: Unique identifier for the customer who made the transaction.

3.product_id
Description: Unique identifier for the purchased product.

4.transaction_date
Description: Date and time of the transaction.

5.quantity
Description: Number of units of the product purchased.

6.total_amount
Description: Total amount of the transaction in local currency.

payment_method
7.Description: Payment method used (card, cash, bank transfer, etc.).

8.store_location
Description: Location of the store where the transaction took place.

Business logic
1. conversion_rate : transaction_attempt / transaction_id

2. Etc....

Abboud Hammour
·
·
Yes I think that would be good (maybe add one line after business logic "this contains business logic for calculations of columns across tables")
Youcef Zeghiche
·
·
Oh thanks ! I have a last question : How Can I help dust to anticipate unpredicted user inputs like "give me transformation rate ?" instead of "conversion rate" ?
Youcef Zeghiche
·
·
I'm sure that i will have many semantic problems
Abboud Hammour
·
·
No perfect science unfortunately other than putting common terms in the instructions, but LLMs are pretty good at guessing even without so I'd test a few!
Youcef Zeghiche
·
·
Okay thank you very much !