Gen AI for Qlik Sense Development - A Master Class

Let's start to answer what a prompt is. At the same time we'll explore some samples to get you into a mindset for prompt construction.

LLM's are trained on a large amount of data contained in the corpus of the Internet as such it has learned patterns and our prompts will influence how it produces interesting information,facts,hallucinations etc.

Not necessarily software, we can give ChatGPT rules to follow instructions to generate output based on your prompt.

Patterns significantly influence the behaviour of LLM's

The Persona pattern is one of the most powerful patterns for tapping into LLM behaviour

LLM's are trained up to a specific date, however we can add more data to help it past that date

ChatGPT has a input limit , as such we have to think about how we design the prompt!

Don't think of prompts as a one off question or single statement but a tool for conversations

Setting boundaries to constrain output to specifics

The LLM may be able to infer patterns and make significant improvements to our question(s)

Breaking the problem down into smaller problems for a better outcome

The LLM can be set to output to a specific audience type

Have the LLM to ask us questions to guide the conversation to a final solution

Teaching the LLM to follow a pattern in the conversation

Teaching the LLM to plan and learn the next action etc

Using Intermediate steps to achieve a goal during the conversation

Making sure we write a good few shot prompt that the LLM will understand

What is this and should I care about it ? For sure you should !

A strategy aimed at facilitating and illustrating the model's reasoning process, we used to have to do this at school in Maths, remember those times :) ? ChatGPT is evolving this process too, we explore it here.

ChatGPT's Flowchart-to-Code capability (Python script generation)

Will ChatGPT help with Qlik Sense coding ?

There is a need for maintaining and evaluating prompts over time, especially when the underlying model or data changes!

Teaching the Model to use formatting rules for output.

Using shorthand to converse with the model

Getting LLM's to fill in the blanks for us. NO not how to bake a cake from a recipe :)

Asking a LLM to define the rules of the game to play with you, this is a very good method to test your ability in Prompt Engineering or other subject domains

Brainstorming to arrive at new approaches to engineer a solution to a given problem.

Make sure the LLM waits for you before it gets' carried away with output.

Did you notice I have been combining patterns in the prompts ?

Helping the LLM to not lose the plot during a long conversation

The Outline Expansion Pattern is a strategic approach designed to overcome the input and output limitations of large language models like ChatGPT.

Defining a set of actions that we can run to avoid repeating long prompts

Avoiding the mistakes and hallucinations output from a LLM

Filter/Replace/Remove text/data from input/output/files

Your thoughts and philosophies will influence the way you interact with large language models.

How do we start a conversation with the Large Language Model that will form the basis for potentially a lengthy conversation to assist us with our project goals?

Getting started in the project with our first prompt

What is a GPT ? How do we create one ? Just watch this, it's like writing a job description and applying it the your GPT!

More about the GPT etc

Customising the GPT to be more efficient and understanding

Generate questions ready to ask our stakeholders so we can start work on the project

Our data resides in a special type of data repository

Final part of the discussions resulting in a data summary document

Ask the GPT to explain a topic you may not know about.

Review the GPT explanation

To get the data into Qlik Sense and ask ChatGPT to review it we'll need to install PostgreSQL

We need the raw data restored to our PostgreSQL local server

Use ChatGPT create a plan to extract sample data for evaluation and

generate the SQL code ready to execute.

Dissect the prompt in your GPT

Here's a tricky one.

Demonstration of features.

Using the AI generated SQL to extract sample table data from the ODS

Testing your knowledge

Confirming your knowledge

What we are trying to achieve in this section

Now that we have extracted sample data we need to understand what it is.

Documentation is very important even if the business does not give us any, we can make our own!

Analyse a sample dataset , document it.

Analysis Demonstration

Get a basic grip on how databases are modelled

Review and understand this multi-step prompt

The GPT analyses two data table files to determine and report significant information and draw the first components of the entity relationships.

Test the ERD against the real data source using GPT generated test queries

Add a 3rd table to the model, watch the lecture even if you do not intend to do the exercise as there is extra tips here about managing chat sessions.

A dialogue is required to remind the chat that it has missed some items.

Solution final including SQL tests

So far we have only analysed 3 tables. Now let's bulk this up!

After much testing I finally got the desired outputs for the project

Our GPT generated PostgreSQL test queries to verify the cardinality of the model , but are they any good ?

Making discoveries during testing that assist in enhancing the overall model

How do we approach fixing the model to include orphaned tables ?

There were 2 orphaned tables, can you fix the second one ?

Solution demo

The learning journey that resulted in the ERD development

Evolution of the prompt to generate the final ERD

And how to prove the 1:1 cardinality

What the final prompt is made of !