Talking to corporate data in a human way
"Empower your customers with personal assistants" - that's the simple but powerful idea behind Kaia's Team, an AI powered SaaS platform made by MAVENS, the leading MarTech engineer based in Hamburg, Germany.
Human-like conversations are the future of customer interactions—multimodal and real-time. Customer Assistants speaks & listen to provide tailored information and execute any task you wish.
Powerful features, ease of use for employees and customers, high security standards and reliability - these are the requirements that we ourselves place on our software. And here we are: It's really easy to bring Kaia into your own team. You only select your foundational model (Open AI or Azure AI), personalize it with your custom knowledge, and effortlessly deploy your customer agent across websites, apps and messengers.
Benefits of AI Customer Assistants
They are 24/7 available. That is not 38h, but 168h per week. No matter whether at weekend, at night or on holidays.
They are cheaper. Our entry tier offers 1.000 chats for 299 Euro/month . That's less than 30ct for an entire customer conversation.
They are unlimited. This team knows no overload. Peak loads and shortages are a thing of the past. At any time. In real time.
What technology uses AI Customer Assistants?
Most popular Customer Assistants - like Kaia's team - are based on Open AI’s latest AI Models and uses their Large Language Models (LLMs - not to be confused with natural language processing - NLP) GPT (and this stands for: "Generative Pretrained Transformer") To make the magic happen: talking to data in a human way.
Basically, LLMs are mapping text to text. Given an input string of text, a LLM predicts the text that should come next. The models we use are so powerful today, that prediction errors are mostly a thing of the past.
Kaia’s LLMs know:
how to spell
how grammar works
how to paraphrase
how to answer questions
how to hold a conversation
how to write and talk in many languages
how to analyze pictures and documents
and many more
Thy do this by “understanding” a large amount of existing text and learning how words tend to appear in context with other words. The model then uses what it has learned to predict the next most likely word that might appear in response to a user request, and each subsequent word after that.
Creating Customers Assistants
Assistants are capable of performing tasks for users. These assistants operate based on the instructions embedded within the context window of the model. They also usually have access to tools which allows the assistants to perform more complex tasks like analysing images or documents or retrieving information from a file.
They can access multiple tools in parallel. These can be tools hosted by our LLM - such as code_interpreter and file_search - as well as tools created/hosted by you (via function calls).
Additionally, assistants can access persistent threads. Threads simplify AI application development by storing message history and truncating when the conversation becomes too long for the model's context length. You create a thread once and simply attach messages to it when your users respond.
Wizards can access files in different formats - either as part of their creation or as part of threads between wizards and users. When using tools, wizards can also create files (e.g. images, spreadsheets, etc.) and quote files that they reference in the messages they create.
Interfaces
Kaia's Team kommt standardmäßig mit einem Website-Widget. Das ist eine kleine Anwendung, die leicht in jede Webseite oder mobile App eingebettet werden kann. Es ermöglicht nahtlose Interaktionen mit Nutzern, indem es direkt auf der Seite angezeigt wird, um Dienstleistungen wie Kundenunterstützung oder Informationsabruf zur Verfügung zu stellen.
You will receive an embed script from Kaia after configuration. Insert the script into the HTML code of your web page, ideally before the closing </body> tag.
Adjust configuration: If necessary, adjust the settings of the widget in your Kaia dashboard.
This is how it looks like
<div config="668fxxxxxxxxxxxxx988297f" host="https://storage.mavens.com" 'id='bl-chat-widget-bubble-button' > </div> <script type="text/javascript" id="bl-chat-widget-bubble-loader" async defer src='https://static-staging-wjs.mavens.com/llm-widget/out.js' > </script>
In addition to the widget, Kaia's team also offers the use of WhatsApp as an interaction channel.
Instructions
Write your instruction in the admin panel of our tool, and the model will do its best to follow the instruction and then stop. Instructions can be detailed, so don't be afraid to write a paragraph explicitly detailing the output you want, just stay aware of how many tokens the model can process.
In general, it is possible to freely define instructions. However, our instruction wizard is happy to help you create them with a more advanced structure. We recommend to mention topics like:
General Introduction
Role Clarification
Overview of Problem and Goal
Restrictions:"Never mention any sensitive or unnecessary information."
General Communication Guidelines
Interactive Approach
Language Flexibility
Response Structure
Repetition Avoidance
Empathy and Simplicity:
Personalization:
Task-Specific Instructions
Function Utilization: "When required, use the 'searchDocument' from Tools to find specific information on products and recommendations."
Issue Identification: "Independently identify issues or problems based on user inputs without asking back. For product-related queries, provide information on similar items, alternative options, and related recommendations."
Advanced Actions
User Response and Engagement
Feedback and Recommendation
Direct Feedback
Proactive Support
Closure and Additional Help
Information Retrieval
User Prompts
How to control a conversation with an assistant: Of all the inputs, by far the most influential is the user prompt.
With an initial instruction produce output in a few ways:
Task: Tell your assistant what you want them to do
Completion: Induce the assistant to complete the beginning of what you want
Scenario: Give the model a situation to play out
Demonstration: Show what you want, with either:
A few examples in the prompt
Many hundreds or thousands of examples in a fine-tuning training dataset
Example instruction prompt:
“Please name only the prime numbers of these numbers separated by commas: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20
Output:
“Here are the prime numbers separated by commas: 2, 3, 5, 7, 11, 13, 17, 19”
Completion prompt example
Completion-style prompts take advantage of how large language models try to write text they think is most likely to come next.
Example completion prompt
please complete the series for the first 20 prime numbers separated by commas: 2, 3, 5, 7, 11, 13, 17, 19...?
Output
Continuing the series, here are the first 20 prime numbers: 2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71
Scenario prompt example
Giving the model a scenario to follow or role to play out can be helpful for complex queries or when seeking imaginative responses. When using a hypothetical prompt, you set up a situation, problem, or story, and then ask the model to respond as if it were a character in that scenario or an expert on the topic.
Example scenario prompt
Your task is to explain prime numbers to a student so simply that he or she can understand them intuitively
Output
Prime numbers are numbers that can only be divided evenly by 1 and themselves. For example, 2, 3, 5, and 7 are prime because no other numbers divide them evenly. They are like the "building blocks" of all other numbers. If a number can be divided by something other than 1 and itself (like 4 can be divided by 2), it's not prime.
Functions on scale
Functions make it possible to perform specialised tasks by allowing the model to access external tools or data in a targeted manner. Thanks to these built-in functions, interaction with users becomes easy for you and your customers. Start surveys, reviews, appointments, bookings, product enquiries - or help them getting explained documents like invoices, analyses, manuals and many more.These functions extend the capabilities beyond pure text generation and make it more versatile in use.
Enabling assistants to fetch data: an AI assistant needs to fetch the latest customer data from an internal system when a user asks “how much did I pay?” before it can generate the response to the user
Enabling assistants to take actions: an AI assistant needs to schedule meetings based on user preferences and calendar availability.
Enabling assistants to perform computation: a math tutor assistant needs to perform a math computation.
Building rich workflows: a data extraction pipeline that fetches raw text, then converts it to structured data and saves it in a database.
Modifying your applications' UI: you can use function calls that update the UI based on user input, for example, rendering a pin on a map.
Image Recognition
With image recognition in Kaia’s Team, users can upload images and ask specific questions about the content of the image. For example, you can upload a photo and ask for details such as describing objects, analysing scenes or identifying text in the image. This function is particularly useful for extracting additional information from images or for providing visual support during conversations. By combining text and image processing in one dialogue, ChatGPT offers a comprehensive way of interacting with different media.
File Recognition
With the file upload function in ChatGPT, users can upload different types of files to get detailed analyses or answers. Whether they are text documents, spreadsheets, PDFs or other file formats, users can analyse the content of uploaded files, search for specific information or even perform complex data calculations. This feature is particularly useful when you want to efficiently process large amounts of information or structured data without having to enter everything manually. ChatGPT can thus provide valuable support when working with large documents and data sets.
Data Analysis
Data analysis with Kaia’s Team enables users to efficiently process and interpret large and complex data sets. Whether it is tabular data, statistical information or text data, Kaia can recognise patterns, show correlations and present results in an understandable way. This function is particularly useful for creating reports, creating diagrams or drawing conclusions from extensive data sources. This enables users to make informed decisions and optimise the use of their data.
Appointments
With the ‘Make an appointment’ function in ChatGPT, users can easily plan and manage appointments. For example, they can specify the date, time and occasion of the appointment, and ChatGPT will help organise the appointment, set reminders or even formulate invitations. This feature is particularly useful for planning meetings, face-to-face meetings or other important events, as it simplifies the process and ensures that all details are clear and well organised. This allows users to keep track of their appointments and ensure they don't miss anything.
Structured Outputs
Structured Outputs is a feature that ensures your assistant always generates responses that align with your provided Schema. This eliminates concerns about missing required keys or incorrect enum values.
Benefits of Structured Outputs:
Reliable type-safety: Ensures responses are correctly formatted, reducing the need for validation or retries.
Explicit refusals: Safety-based refusals are now detectable by your system.
Simpler prompting: Achieve consistent formatting without complex prompts.
Fine-tuning
LLMs are pre-trained on extensive text data. Typically, prompts include instructions and examples, a method known as "few-shot learning."
Fine-tuning enhances the models available through the API by offering:
Higher quality results compared to standard prompting.
Training on more examples than a single prompt can handle.
Token savings with shorter prompts.
Lower latency** in requests.
Fine-tuning goes beyond few-shot learning by training on a larger set of examples, delivering better results across a range of tasks. Once fine-tuned, the model requires fewer examples in prompts, reducing costs and improving response times.
Unlock real time assistants
Kaia's Team integrates well with your tech stack: connected to Hubspot, Salesfore, Zapier, Gmail, and others. With our API, you have over 5,000 real-time applications at your disposal.
Some of the most popular apps you can use
Sum up
Customer Assistants are revolutionising customer interactions by using the latest AI models for highly personalised interactions. Instead of rigid workflows, they offer flexible conversation modules that fully adapt to the user's needs. The platforms are user-friendly, easily customisable and optimised for comprehensive support. They offer advanced capabilities such as real-time data integration, API connectivity and seamless transitions to human agents when needed.
Solutions like Kaia's Team improve the customer experience dramatically, help to reduce costs and are suitable for any industry.
Comments