Moving Beyond Custom Consumer-Facing AI Chatbots & Getting Insights From Your Businesses Big Data…

Orren Prunckun
2 min readMay 24


Back in February, I wrote ( a free Google Sheets script that would allow people to bypass trying to write a formula (or complex formula) for existing data and it would ask OpenAI to do the heavy lifting and provide an answer/output without needing a formula.

That is still free if anyone wants it, but it only works for a single spreadsheet.

And OpenAIs gpt-3.5-turbo model can only take just over 4000 tokens (read that simply as approximately 4000 spreadsheet cells).

So, what if you have lots of spreadsheets?

You can upload each spreadsheet into a web app and then run that uploaded spreadsheet through OpenAIs gpt-3.5-turbo model for a natural language response:

A web app would also allow you the liberty to add a chat feature that can allow you to ask follow-up questions or different questions of the data to start with.

What if your data is bigger than 4000 spreadsheet cells (i.e. 4 columns with 1000 rows for example)?

To overcome this limitation, you can create what is called a Semantic Search engine…

You’d split all the different sub-data (i.e. each spreadsheet or spreadsheet tab) into a chunk of a maximum of 4,000 cells, then you’d search your input prompts against each of these chunks and only return those relevant chunks from your database to finally analyse via OpenAIs gpt-3.5-turbo model.



Orren Prunckun

Entrepreneur. Australia Day Citizen of the Year for Unley. Recognised in the Top 50 Australian Startup Influencers.


See more recommendations