CopiLot / Judi - AI - LLM (Alpha)

This page introduces the AI capabilities within pi. Now we are not 100% sure what we will call this… but this is a good place to start. In this document I will call it the AI for now.

Prerequisites / Getting Started

The first version - links to OpenAI - so you will need an OpenAI account and generate an application TOKEN.

To generate an OpenAI API token, follow these steps:

Have a demo environment: In this initial test phase we expect you to use this for testing only, and have this built into an environment with test data.
Licence Key: You will need a piPredict module capability within your licence.
Sign Up or Log In: Visit the OpenAI API Platform and sign up for an account or log in if you already have one.
Navigate to API Keys: Once logged in, click on your profile icon in the top-right corner and select "API Keys" from the dropdown menu.
Create a New API Key: Click the "Create new secret key" button. A dialog will appear displaying your new API key.
Secure Your API Key: Copy and store the API key in a secure location immediately, as it will not be displayed again.

For detailed information, refer to OpenAI's official documentation on API keys.

Ensure you keep your API key confidential and avoid exposing it in public repositories or client-side code.

Add the key into the pi Variable: On the System Setting screen, there is a new section for LLM. This is where you control the API keys from your LLM’s

Refresh you browser: You should now see a small flask icon in the bottom right hand corner of your charts.

Click the Flask on a chart you’d like to know more about: The LLM will produce a summary, Insights into the data and suggest actions you may wish to consider.

For Example: The AI looks at this table - of data. It is from a Consumer Duty dashboard, and looks at how happy our customers are based on the Products and Services we offer, whether they think it provides fair value and how well they understand how to access the services.

The AI then generates text like this;

Summary:

- The chart shows the percentage of targets met by different areas within the company based on survey data. - The areas analyzed are Products and Services, Fair Value, and Consumer Understanding.

- The data is broken down by product category: Motor, Home, Travel, and Pet.

Insights:

1. Motor and Home products have met or exceeded the targets in all three areas, with Home products performing exceptionally well across the board.

2. Travel products have met the target for Fair Value but fall short in Consumer Understanding.

3. Pet products are struggling in all areas, particularly in Consumer Understanding where they have not met the target.

Actions:

1. For Travel products, focus on improving consumer understanding through clearer communication and education about the product features and benefits.

2. For Pet products, a comprehensive review is needed to enhance both fair value and consumer understanding. Consider revising the product offering or providing additional training to staff to improve customer knowledge.

3. Recognize and reward the teams responsible for Motor and Home products for consistently meeting or exceeding targets in all areas. Encourage knowledge sharing and best practices across different product categories.

Not bad hey!!!

Warning

⚠️ Warning Notice for ChatGPT API Usage

Important Notice:

When you interact with this application and send a request to the ChatGPT API, please be aware of the following:

Data Transmission:
- The data you enter will be sent to OpenAI's ChatGPT API for processing.
- This may include any text, queries, or information you provide.
AI-Generated Responses:
- Responses are generated by an AI language model (LLM) and are based on patterns learned from vast datasets.
- Results may not always be accurate, up-to-date, or contextually appropriate.
Verification Required:
- Do not rely solely on the responses for critical decisions, legal matters, medical advice, or financial guidance.
- Always verify information from trusted sources before taking action.
Data Sensitivity:
- Avoid sharing personal, confidential, or sensitive information when interacting with the chatbot.
- This application does not store your chat history beyond the session.

Chatbot

Generate Report

One of the most useful features we have found so far is to generate a report with the AI providing the same context about the data.

Note: This is manual at present but will become more integrated - it will shortly allow you to generate / schedule a report with the AI context.

Let’s walk through creating a report over the pi example Consumer Duty dashboard. To give you (the reader) some context to this; The Consumer Duty dashboard is designed to be used by organisations regulated by the FCA. It is important that if you are sold a loan, insurance etc, that;

The product(s) and services are appropriate to you: (you were not mis sold / over sold - for instance if you needed to make a claim - did it cover what you expected),

That the products offer fair value (you were not ripped off, is the company exploiting people and charging too much)

Do you understand how to access the services (If you bought an insurance product and can’t make a claim when you need then it has let you down)

How good is the support? (Do you just end up in a call queue, or do the digital services help solve problems you may have.)

To make sure that a company is full filling it’s requirements they will track data from their premium applications, take surveys with customers and monitor telemetry systems.

Check out this link ---- KEN ---- to more information about the consumer duty dashboard if you are interested.

So I’ll Start to create a report as Normal;

Including charts, let’s start with the summary;

Explain Chart

When I add a chart - I can now generate the text for this chart - or select any on the page, simply by clicking the Flask icon in the new Generate Text box.

Clicking the Flask asks the AI to generate the text for the chart, now I can copy and format the text to a chart cell on the report (this will become automatic)

If I generate a pdf from the report, it shows the pi element, and the AI generated text that I have placed on the report.

When I have multiple charts on a page, I can choose which chart to generate this for.

I can continue to add the text for nay charts that I want to provide this description for.

Explain a Page

Coming soon….. This looks at the data in all the charts on the page and provides an overall page level summary view.

Explain the Report

Coming soon….. This looks at the data in all the charts in the whole report and provides an overall report view.

Future

Pi will soon allow secure integration - into any LLM including private / closed loop LLM's. Pi will not only allow you to not only secure the data by user, role, tenant and organisation, but also pass data context to the LLM to reduce the level of hallucination. In addition, pi will allow you to create workflows with connectors to all major models + model providers, letting you test models side by side and gain accuracy feedback from users. It will offer the abilities to link to best of breed workflow engines such as Langfuse - to allow RAG and Agent creation.