Download the case study

SolvPath: Scalable Gen AI-Powered Platform for E-Commerce Query Resolution

Category: E-commerce

Services: Gen AI Development, Cloud Architecture Design and Review, Managed Engineering Teams

95%

more accuracy in understanding user queries by leveraging Amazon Bedrock

90%

accuracy and contextually relevant responses to user queries using RAG

99.9%

uptime and real-time user interaction using Amazon OpenSearch

85%

85% enhancement in the chatbot’s semantic understanding by integrating Amazon Titan

60%

60% reduction in manual customer support ticket resolution

About SolvPath

SolvPath is an advanced AI-driven customer support interface designed to assist E-commerce platforms by delivering accurate, real-time answers to customer queries. It utilizes LLM capabilities to enable seamless natural language understanding, improving customer experience through contextual and accurate responses.

Challenge

Accurately understanding diverse customer queries, including those with spelling errors or complex sentence structures, and providing contextually correct answers.
Ensuring the assistant is available at all times for real-time interaction without compromising performance.
Enhancing the assistant’s semantic abilities to understand the deeper intent behind queries and deliver more personalized answers.
Designing a solution that scales efficiently to handle increasing user queries without downtime or performance degradation.
Ensuring the secure storage and handling of sensitive customer data while processing vast amounts of user information.

Solution

Query processing integration: We integrated Amazon Bedrock to enhance SolvPath’s ability to interpret and respond to complex and varied customer queries with high accuracy.
Real-time response system: A setup using Amazon OpenSearch was established to ensure seamless, low-latency customer interactions by processing data in real time.
Retrieval-augmented generation setup: We incorporated Retrieval-Augmented Generation (RAG) with Amazon OpenSearch and Bedrock to generate accurate, reference-backed responses by accessing a vast knowledge base.
Semantic understanding enhancement: We configured Amazon Titan to improve the platform’s ability to understand the intent behind user queries, allowing for more personalized responses.
Model deployment and management structure: We leveraged Amazon Bedrock to streamline the deployment and management of large language models, ensuring smooth scaling as needed.
Secure data storage implementation: We established a scalable, secure storage solution using Amazon S3 to manage large volumes of data and user information effectively.
Scalable infrastructure setup: An infrastructure was built with Amazon ECS and AWS Fargate to ensure efficient scaling in response to increasing user demand while maintaining high performance.
Monitoring and optimization system: A monitoring system was integrated using LangSmith to track performance metrics and enable continuous optimization of the platform.

Outcome

95% accuracy: Achieved high accuracy in understanding and processing complex customer queries.
90% contextual relevance: Responses provided were contextually relevant, improving customer satisfaction.
60% reduction: Reduced manual resolution of customer tickets by 60%, improving operational efficiency.
85% improved understanding: Enhanced semantic understanding, improving the assistant’s ability to deliver contextually rich answers.
99.9% uptime: Ensured continuous availability of the AI-powered interface, supporting real-time customer interactions.

Architecture Diagram

AWS Services

Amazon Bedrock: We used Amazon Bedrock as a foundational framework for building and deploying AI-powered applications. It also ensures AI models’ reliability, scalability, and performance and facilitates seamless integration with other AWS services.
Amazon Bedrock Claude-2: Our experts used Bedrock Claude-2 as the core language model for the chatbot to ensure it generates contextually relevant responses to user queries. It helps understand nuances in user queries and provides accurate information.
Amazon Titan: We used Amazon Titan to enhance the chatbot’s semantic understanding and generate more context-aware responses. It also helps the chatbot understand the intent behind user queries, leading to accurate and relevant responses.
Amazon S3: We used Amazon S3 to provide a scalable and secure storage solution for vital chatbot data such as FAQ databases and training data.
Amazon OpenSearch: Our developers used Amazon OpenSearch to deploy and manage large language models like Claude-2 easily. We also used OpenSearch to enable real-time user interaction and provide timely and accurate responses to users.
Amazon EC2: Our AWS experts used EC2 to host the chatbot app and provide computing power to process user queries and generate relevant responses.
AWS Fargate: We used AWS Fargate to run and deploy the chatbot application as a container to ensure scalability and efficiency in resource allocation.
Amazon ECS: Our developers used Amazon ECS to manage and scale containerized components of the AI-powered chatbot ecosystem.
Amazon RDS (PostgreSQL): We used RDS to store structured data related to a chatbot, such as user queries, interactions, logs, and feedback, to improve performance.