![• Max Number of tokens to generate
should be 4096.
• You need to add regex in Dynamic
Prompt Generation.
• As there is backend code to mask if User
Entered value is more than 10 digit. So
you can add a regex for it as “[0-9]{40}”
and for email validation you can add “^[
w-.]+@([w-]+.)+[w-]{2,4}$”
• CSS File Link : Link](https://image.slidesharecdn.com/llmcommerceframework-250101142641-5df03ba2/85/rich-media-with-buttons-LLM-Commerce-Framework-pptx-11-320.jpg)
rich media with buttons LLM Commerce Framework.pptx
- 1. LLM Commerce Framework Prepared By – Vishwarajsinh Jadeja Prepared for - Streebo
- 2. Table of Content • What is LLM Wrapper • Why to use LLM and what are its benefits • Process flow diagram of LLM • Prerequisites • When to Use LLM Commerce Framework • Types of Websites • Requirements for LLM Commerce Framework and Configuration of Config Form • What are Steps For Integration LLM Commerce Framework in our Bot • What to add in Our LLM Prompt • Requirements for Buy Product • Requirements of Track Order • Requirements of Cancel order • RCA and Debugging • Exercise • Solution video and prompt for the exercise
- 3. What is LLM Wrapper In LLM Wrapper, we are developing our chatbot using only LLM prompts. This approach ensures that every aspect of the chatbot's functionality is driven by these prompts. Specifically, we handle all transactional flows, frequently asked questions (FAQs), irrelevant questions, and out-of-domain inquiries using LLM prompts alone. This method allows us to create a seamless and efficient chatbot experience without relying on additional scripting or external systems.
- 4. Why to use to LLM and what are its Benefits • Using an LLM Wrapper can make developing Proof of Concepts (POCs) much faster and easier. Instead of having to write a lot of code and manage different fields like you would with Leap, you can handle all the steps for transactional use cases and validations directly in the LLM Wrapper. This saves time, reduces complexity, and allows you to focus on refining your POC without getting stuck in complex coding.
- 5. Process Flow Diagram For LLM
- 6. Prerequisite s • Good Understanding of Prompt engineering – For mastering the prompt engineering you can refer below short course. • Course Link: https://www.deeplearning.ai/short-courses/chatgpt-prom pt-engineering-for-developers/ • Learn about LLMs: Understand how large language models (LLMs) work, their architecture, and their limitations. Get friendly with LLMs like ChatGPT, Gemini. • Good Understanding of Config Form • Label Binding Understanding • Study NLP: Gain a solid foundation in natural language processing (NLP) concepts, including tokenization, embeddings, and model training.
- 7. When to Use LLM Commerce Framework • As there are two type of Website as Content based Website and Product based Website. • Content based website are like Wikipedia, primarily provide information and Flipkart, amazon that all are Product website. • So, our LLM Commerce Framework we will use it for Product based website.
- 8. Types of Websites 1. Content Website
- 10. Requirements For LLM Commerce Framework and Configuration of Config Form •Include a CSS file to style the HTML rendered by GPT. •Ensure the Config Form is filled out. •Turn off Cog search and NLP functionalities. •Enable both checkboxes: "Generate dynamic prompts using AI Engine" and "Process Irrelevant Intent.“ •Write the prompt for the LLM Wrapper. •Utilize either GPT-4o (With Permission from your Manager) or GPT-4mini.
- 11. • Max Number of tokens to generate should be 4096. • You need to add regex in Dynamic Prompt Generation. • As there is backend code to mask if User Entered value is more than 10 digit. So you can add a regex for it as “[0-9]{40}” and for email validation you can add “^[ w-.]+@([w-]+.)+[w-]{2,4}$” • CSS File Link : Link
- 12. How to disable NLP in Config Form. • Go To Config form and There Go to NLP Engine Tab and Select None from it.
- 13. How to Disable Cognitive Search in Config Form • Go To Config form and There Go to Search Engine Settings Tab and Select None from it
- 14. Where to Add LLM Wrapper Prompt in Config Form • Go to Your Config Form and there go to Dynamic Prompt Generation Setting and Enable “Do you want to process Irrelevant Intent?” and “Do you want to generate dynamic prompts using AI Engine?” and add Your Prompt in “Conversation description” and also add your regex in regex Field Like Shown Below.
- 16. From Where to Change GPT Model and Increase Number of tokenizer to Generate in Config Form • In Your Config Form Go to Dynamic Prompt Generation Setting Tab and There Scroll Down and You will find a section of OPEN AI Creds as shown below from there you can change GPT Model and Increase tokens to generate.
- 18. What are Steps For Integration LLM Commerce Framework in our Bot. • First Need to Write a LLM Prompt with a HTML Template that will help to show product carousel and need to need to add a product catalogue where we have add our product information like image url, product name, product price etc based on our requirement. • Need to use gpt-4o (With Permission from your Manager) or gpt-40-mini model to show product carousel based on the HTML you provided in the Prompt. • Need to Add a CSS file in our theme app to give a style to the HTML provided in LLM Prompt.
- 19. What to add in Our LLM Prompt? • You Need to train the prompt for the specific domain and you need to add a keywords as per your domain. So that the user’s question doesn’t consider under out of domain. • You need to also add extra information in your prompt like Contact Information, Team Information, Location etc. • You need to add HTML Template in Your Buy Product Prompt for rendering Product Carousel. • HTML Template Prompt Link : Link • You need to add all the transactional Usecase that you want in our bot. • Transactional Usecase Can be : 1 Buy Product, Track Order, Modify Order, Cancel Order etc for ecommerce websites.
- 20. • You need to add a Product Catalogue where there are product details like Product Name, Product Price, Product Description, Product Image etc as per your requirement – Prompt Link
- 21. Requirements for Buy Product • If user’s query is about buy products then first thing you need to allow user to select category and based on your category select if there are subcategory then allow user to select subcategory based on category selection. • Once user has selected category and subcategory You need to add a html template that will be render by adding style based on subcategory selection. • Once Product carousel is visible to user then we will also give add button in each product so user can click on it and can add that product in there cart.
- 22. • Once user has added product in there cart. We will give prompt user that they want to continue shopping or proceed to checkout or View there cart. • If user has selected to continue shopping then we will go to buy product flow again to allow user to select category and subcategory and based on it we will render the html and allow user to add product and give him option what user want to do further. • If user has selected view cart then we will show product details with price to know user that these are product added and to know its worth.
- 23. • If user has selected Proceed to checkout then we will ask basic details from user like full name, mobile number, address, card details like card number, CVV and expiration date and once user has all the details we will ask OTP from user that will be fix like OTP will be “252019” that we have added in the prompt that if user gives this OTP then only go further otherwise ask OTP again and once user has provided correct OTP then provide order number to user of any 5 digit. • Buy Product Prompt Link : Link
- 26. Requirements of Track Order • For track order we will ask order number from user and based on order number will give prompt to user that your order is delivered and it will receive you by this date and it is in this location right now. • Track Order Prompt Link : Link
- 28. Requirements of Cancel order • For Cancel Order We can ask Order number from the user based on it we will ask cancel reason from the reason and after that we will ask OTP from user that will be fix like OTP will be “252019” that we have added in the prompt that if user gives this OTP then only go further otherwise ask OTP again and once user has provided correct OTP and then we give a prompt that our order is cancelled and we will sent you SMS to your registered mobile number. • Cancel Order Prompt Link : Link
- 30. RCA and Debugging • http://dxasolutionsdev.streebo.com:9002/lognavigator/logs/chatbot/list?subPath=chatbot- backend • http://dxaacademy.streebo.com:9002/lognavigator/logs/chatbot/list?subPath=chatbot- backend • Username :: strlogs • Password :: strlogs
- 31. • You can check the Logs Regarding LLM Wrapper by Using Specific Environment URL. • Link: LogNavigator (streebo.com) • Now Using Above link, you will get logs list. You can check logs in service. • You have to find the Config Key and You can check it date wise. • You should check the logs if any error occurred or bot not giving expected answer. • Error Comes in logs:- • Solution: Go to your config file, add correct API key as per environment. You can take the API key from your TL/Manager.
- 32. Exercise • Exercise 1 Create a Transactional Bot That Can Help user to buy product, Track order and supports Rich Media as Bot can Show product carousel based on category selection by the user. You Can create Bot For this Website: Website Link • Exercise 2 Create a Transactional Bot That Can Help user to buy product, Track order and supports Rich Media as Bot can Show product carousel based on category selection by the user. You Can create Bot For this Website: Website Line
- 33. Solution video and Prompt for the exercise • I have created a one bot for Burger King. You can refer video from below and create a bot for solutions mention in exercise. • Recording Link: Link • Prompt Link: Link