In this article, DataArt’s team talks about a GPT-powered solution for accomplishing do-it-yourself (DIY) projects, which they built in collaboration with marenas, a Munich-based business consulting company. They invite you to check out Buy-me-Bot* in action and see how the GenAI system reshapes customer interactions.
Basic Capabilities of GenAI Systems
GenAI excels in comprehending queries expressed in a natural language, providing a seamless user experience. It converts queries into a searchable format, tapping into extensive knowledge bases to retrieve relevant information. GenAI not only answers questions but also provides detailed analyses, offering customers comprehensive support. Customer queries are addressed round the clock and within seconds, enhancing the overall service availability. The system is able to recognize its limitations and involves human support when needed, ensuring a balanced and effective support system.
Automated DIY Project Support with Buy-me-Bot
DataArt’s team of AI&ML experts developed an interactive bot for creating DIY projects. The bot can compile a list of essential materials and a step-by-step DIY manual and provide users with three price-ranked bundles of required items and the links to order them on eBay. It leverages Large Language Models (LLMs) as an engine to instruct and recommend how to build DIY items. In addition, the bot can be tuned and integrated with various e-commerce platforms to give users the best bundles of materials to buy.
Buy-me-Bot serves as an example of fast prototyping and experimentation from both organizational and technical perspectives, which is so much encouraged by DataArt’s Labs– the solution was completed in slightly over one month.
From an organizational perspective, we saw a potential market niche, with an interest in using AI/ML technologies to develop solutions allowing customers to assemble their products upon request. We swiftly assembled a compact team, involved DataArt's R&D budget, and quickly implemented a Proof of Concept (PoC).
Following this, we showcased the solution to potential customers and received their feedback. This gave us grounds to believe that the product may hold interest for the market and plan for the prospects of its further development, precisely the approach required in today’s rapidly changing world.
From a technical standpoint, we leveraged cloud systems, a trend of recent years, on the one hand, and on the other hand, used Streamlit. This‑ open-source Python library enables machine learning engineers to create diverse web applications easily. This library allows web applications to be published online and also includes a built-in web server that may be deployed within a Docker container. Of course, this library is not intended for creating production systems, but it is an excellent choice as a tool for rapid prototyping and hypothesis testing.
Key Advantages of Buy-me-Bot
- Enhanced customer satisfaction through personalized assistance and simplified shopping experience
- Increased average number of items in the shopping basket
- Cloud-native Azure-hosted solution with the AI model and UI
- Reinforced with Azure Open AI Services and GPT model
- Compatible with most e-commerce platforms (e.g., eBay or Amazon) via API
- Fully secure and compliant
Technical Overview
Let us now dive deeper into the technical details of the solution. Imagine a platform where users can share their project ideas and receive comprehensive instructions to bring their creative visions to life using advanced technologies. This is what its architecture may look like:
1. Processing User Requests with NLP
The process starts with the user submitting their project request in a natural language. We leverage the Azure OpenAI NLP capabilities to recognize multilingual speech, analyze the user input, and extract crucial information about the type of product to be built.
2. Generating a Material List
Based on the instructions, the system generates a list of materials required for the DIY project. The capabilities of the GPT model are used to identify specific materials needed based on the user's instructions and project specifications. This list ensures that users have everything they need to execute their project.
3. Utilizing Azure OpenAI Services for Instruction Generation
Next, the request is forwarded to Azure OpenAI Services. The language model generates clear instructions based on the gathered information. A step-by-step guide is presented to the user, supporting the successful implementation of the DIY project.
4. Procuring Products Using the eBay API
The system can now access the eBay API with the provided material list. It immediately begins searching and provides a list of items available on eBay that match the material requirements. Users can conveniently purchase the necessary items directly through the eBay marketplace. However, it is essential to note that the eBay API can be replaced with a custom e-commerce API tailored to a specific online store or provider, offering users the flexibility to choose the platform for material procurement.
5. Comprehensive Approach to Requesting and Processing Responses from Azure OpenAI
In the process of making requests to Azure OpenAI, a template response was employed to acquire supplementary instructions for formulating a precise answer. When generating the list of materials, the request is composed primarily in English, with some elements integrated in German. After receiving the response, a dedicated schema and a parser developed by LangChain are utilized to transform it into a JSON-formatted output, facilitating the ease of interaction.
Here is an idea of what a ChatCompletion request can look like. The input message can be divided into two segments: system and user content. For system content, it is recommended to frame a message like this: 'You are a helpful assistant that understands <context>.' As for user content, you can structure a message as follows: 'Example of response: <example>. \n <main instructions>' or 'Request: I want to … \n Instructions: <an ordered list of instructions>.' Additionally, a ResponseSchema can be employed in conjunction with StructuredOutputParser to obtain a JSON-formatted output. For instance, a message might be composed as follows: 'Task: <task> \n Output format: <format instructions from schema parser>.’
6. Cloud-Agnostic Architecture
Finally, the solution architecture is designed to be cloud-agnostic, providing the freedom to deploy the system in the preferred cloud infrastructure. Whether it is Amazon Web Services (AWS), Microsoft Azure, Google Cloud, or any other cloud provider, the architecture can adapt to different cloud environments, offering flexibility and scalability.
Insights
The integration of GPT Turbo 3.5 into Buy-me-Bot has allowed the development team to reveal the following insights:
- Multilingual Proficiency. GPT Turbo 3.5 demonstrates impressive multilingual capabilities. While the model is instructed in English, it can integrate other languages. This flexibility is crucial for global projects and ensures the model's broad applicability.
- Structured Outputs. One of the key features of this PoC is the model's ability to provide structured results in tables. This showcases GPT Turbo 3.5's ability to understand and adhere to specific output formats, which can be particularly useful for tasks requiring organized and systematic data representation.
- End-to-End Solution. Buy-me-Bot is not just about getting a list of items. It provides an end-to-end solution to users, from listing required items to generating instructions and sourcing the items through eBay. This approach offers users a comprehensive DIY experience.
- Product Decomposition. The model is able to break down complex tasks or products into their fundamental components. It can differentiate between essential and optional elements, ensuring that users get a thorough understanding of what is critical for their project.
- Making Guides. Apart from sharing materials lists, the model can generate easy implementation steps. It is like having a helpful friend guide you, making new projects feel easy to deliver.
- Teaming Up with eBay. By linking up with the eBay API, the model shows it is not just smart but also practical. Instead of just giving ideas, it points you straight to where you can buy what you need, turning those ideas into tangible actions.
Other Use Cases for GenAI Bots
The applications of GenAI solutions can be diverse. For example, access to purchasing, production, and logistics systems ensures accurate and real-time delivery information. Resolving issues like ‘my smartphone’s not turning on’ involves accessing detailed product and support information. Finally, if the user inquires about IT security standards implemented in their firm, the system can tap into the company knowledge base and retrieve the required information.
The versatility of GenAI bots extends beyond traditional customer support. Here are some compelling use cases that demonstrate the wide-ranging applications of this technology.
- Recommendation Systems
GenAI powers recommendation systems for product selection, offering personalized suggestions based on the customer knowledge base. - Product Verification
In marketplaces with numerous suppliers, GenAI ensures accuracy by comparing product text descriptions with images. - Cross-Platform Product Search
GenAI can perform fuzzy search, extracting categories from vague requests, alongside advanced cross-based search to provide relevant recommendations across different suppliers. - Visual and Personalized Search
By incorporating visual and personalized search features, GenAI reduces the time to the first purchase on e-commerce platforms. - Goal-Based Product Creation
Users can assemble products based on specific goals, such as creating a birdhouse or establishing an apple orchard. - Customized Fitness and Nutrition Plans
A GenAI bot can create personalized fitness plans with associated nutrition guidance, offering a tailored experience that would come most conveniently on your mobile device.
In summary, GenAI bots have a truly transformative role in customer support. Buy-me-Bot, a cloud-native solution with GPT under its hood, provides a user-friendly experience, enhancing customer satisfaction through personalized assistance. The bot stands out by offering an end-to-end solution to its users, simplifying complex tasks, and teaming up with eBay for practical procurement. In essence, it marries pioneering technology with practical solutions, making it a valuable companion for creative endeavors.
*Feel free to reach out to Darya.Ostapenko@dataart.com to obtain the credentials for Buy-me-Bot.
Co-authors of the article:
Andrey Sadakov, Delivery Manager
Anton Liutov, Senior AI/ML Engineer
Dmitry Baykov, Technical Director AI/ML
