How Winweb Built its AI Assistant with DataStax Astra DB and LangChain

Winweb specializes in ERP software for the food and wholesale sectors. It ensures compliance, streamlines processes, and offers comprehensive functionalities like batch tracking and inventory management.

In response to the growing demand for streamlined client support and the desire to stay ahead in technological advancements, Winweb embarked on an ambitious journey to integrate AI into its customer service infrastructure. Recognizing the potential of AI-driven solutions to improve customer interactions, Winweb set out to develop an AI Assistant.This intelligent assistant is trained to answer questions about Winweb by obtaining information from the WinwebWiki and winweb-food help.

Through the collaboration with DataStax, Winweb introduced a chatbot powered by retrieval-augmented generation (RAG), which enables Winweb to respond precisely to complex queries by performing targeted searches across diverse information sources. The result? Winweb customers gain access to highly relevant and accurate information, significantly enhancing their interaction with the service. 

Technology Stack Overview

Retrieval Augmented Generation (RAG) Flow

Content from platforms such as WinwebWiki and winweb-food help, along with Toggle features, is transformed into embeddings for storage in DataStax Astra DB, converting text into dense vector formats. WinwebWiki data is sourced through GraphQL from PostgreSQL, and functionality toggles are managed through SQL Server queries. Content from the winweb-food help site is also extracted as HTML files, allowing for efficient handling and retrieval of diverse data types in a unified database environment.

A custom service, crafted with Python, Flask, and Docker, facilitates access to this database. Designed to dynamically scale with request volume, it's hosted on Google Cloud Run, with Google Firebase for app hosting, and Google Cloud Storage managing files. This interface handles queries from Winweb apps or winweb-food, executing searches within the vector database to swiftly deliver the most relevant documents, thereby ensuring real-time data access and response.

Winweb is currently using Azure OpenAI's GPT 4 as the LLM and text-embedding-ada-002 as the embedding model. It integrated LangChain to load and retrieve data from Astra DB efficiently. 

Winweb considered other vector solutions including Pinecone but factors such as scalability, performance, and cost influenced their decision to go with Astra DB. 

Safeguarding Customer Data

Prioritizing the most stringent data protection standards, Winweb and DataStax ensure that customer data is secured by employing current encryption protocols and hosting all servers within the European Union. This commitment subjects all data processing activities to the robust EU data protection guidelines.

Results

Winweb’s journey from proof of concept (POC) to full production deployment with Astra DB was accomplished in less than 30 days, showcasing its efficiency and ease of integration. Winweb’s AI Assistant has significantly slashed the time required to find answers from its wiki pages. By leveraging the chatbot, the time required to locate the correct information is expected to be reduced by 20X. Previously, users had to sift through several wiki and help pages to find answers. Now, the chatbot simplifies this by synthesizing data from multiple sources into a singular, coherent response, significantly improving efficiency and user experience.

Additionally, the chatbot is anticipated to decrease the volume of support required for first-level use cases, which typically represent around 10 percent of all cases. This reduction is expected to optimize support workflows and improve overall operational efficiency.

What’s Next?

Winweb’s next project is to develop custom chatbots tailored to each customer's specific domain data, utilizing Langflow to significantly decrease the time required to set up these multiple chatbots. Langflow's streamlined approach to component integration and its array of prebuilt connections and visual tools will enable Winweb to experiment and iterate at an unprecedented pace.

Furthermore, they're looking to implement text2SQL features, enabling customers to interact more effectively with their sales data. This strategy aims to improve customer experience by offering more customized and interactive tools for providing deeper, contextually rich engagements with their own data.

Different embedding models can also encode nuanced meanings in unique ways so Winweb will continue experimenting with a range of models which embeddings  produce the most accurate and relevant results, fine-tuning their systems to excellence. The Astra DB and Langchain framework simplifies the process of transitioning between various embedding models with minimal effort.

Conclusion

The deployment of Winweb’s AI assistant elevates the efficiency and effectiveness of its customer support and sets a new standard for AI-driven client interactions in the industry. By embracing the power of AI, Winweb reaffirms its commitment to delivering exceptional service and staying at the forefront of innovation.

This solution enables businesses to not only manage their operations more efficiently but also to connect with customers in a more meaningful way.

Get started building on Astra DB for free and accelerate your adoption to AI with Langflow