AllCloud Transforms QuesTek’s ICMD Platform Using Generative AI on AWS

By Bharath Terala, Sr Partner Solutions Architect – AWS
By Priten Patel, Account manager – AWS
By Changning Niu, Director of Digital Products – QuesTek
By Jiadong Gong, CTO – QuesTek
By Vincent Chow, Machine Learning Engineer – AllCloud

AllCloud

In a recent joint effort, QuesTek and AllCloud explored the integration of generative AI into the ICMD platform, yielding promising results across several use cases. In this post, we will provide a high-level overview of the generative AI use cases that transformed the ICMD platform.

QuesTek Innovations LLC, a global leader in Integrated Computational Materials Engineering (ICME), has successfully deployed it’s ICMD® (Integrated Computational Materials Design) platform on the Amazon Web Services (AWS) cloud. They did so by leveraging the expertise of AllCloud, an AWS Premier Partner. This move is part of their ongoing collaboration through the AWS Migration Acceleration Program (MAP).

The ICMD platform is a cutting-edge tool that enables scientists and engineers to accelerate the design and development of new materials. Toolkits utilize QuesTek’s Materials by Design® technology to provide proven models, datasets, and workflows, including third-party software and databases, to support design processes. Through the use of advanced simulations and data analysis, ICMD helps clients streamline R&D processes, reduce costs, and improve materials performance.

QuesTek collaborated with the cloud experts at AllCloud to deploy their ICMD platform on AWS. This integration has enabled QuesTek to incorporate several advanced capabilities. This post will dive deeper into each of these use cases:

• Use Case 1: AI-Powered Q&A Assistant: QuesTek used Amazon Lex to develop an AI-powered virtual assistant that can engage in natural language Q&A, helping users quickly find information and get their questions answered.
• Use Case 2: Streamline Simulation Jobs: Using AWS Lambda and other serverless technologies, QuesTek built an intelligent agent that can automatically configure and optimize simulation jobs, streamlining the R&D process.
• Use Case 3: Generative AI Analyzer: QuesTek tapped into the power of large language models on Amazon Bedrock to create a generative AI system that can analyze simulation results and materials data, generating insights and recommendations.

Use Case #1: AI-Powered Q&A Assistant

The first use case involves an AI-powered Q&A virtual assistant designed to assist users by leveraging a comprehensive knowledge base that includes ICMD documentation and extensive background knowledge in materials design and engineering. This technology aims to enhance user integration and streamline access to information, making it easier for users to find answers and solve problems efficiently.

AllCloud leveraged Amazon Lex and Amazon Bedrock Knowledge Bases to implement Retrieval-Augmented Generation (RAG). RAG is a technique that enables large language models (LLMs) to reference document data to answer natural language queries. The ICMD stored product documentation in an Amazon Simple Storage Service (Amazon S3) bucket and had it sync with Amazon Bedrock Knowledge Bases. Further, by taking advantage of AMAZON.QnAIntent in Amazon Lex to securely connect with Anthropic’s Claude Model, a Lex Q&A virtual assistant was built to answer product-related customer questions.

The following figure (Figure 1) shows a Q&A flow described in the first use case with Amazon Lex, Amazon Bedrock Knowledge Bases, and Vector Engine for Amazon OpenSearch Serverless.

Figure 1: High-level architecture diagram of the solution described in use case #1.

Use Case #2: Streamline Simulation Jobs

The second use case features a smart assistant that aids users in setting up complex ICMD simulation jobs for materials design. This assistant not only intelligently makes the setup process straightforward by collecting critical user input data through conversation, but also helps the user generate polished and interactive plots. These features make data visualization more intuitive and informative, helping users understand and interpret their simulation results more effectively.

To accomplish this second use case, AllCloud continued to iterate on the Amazon Lex chatbot/virtual assistant implemented in the first use case. First, a series of custom Amazon Lex intents were implemented using the visual conversation builder for Amazon Lex. As a customer interacts with the chatbot, information (slots) are collected and used to fulfill the intents by using a custom AWS Lambda function attached to the chatbot. Next, the Amazon Lex chatbot was deployed and embedded within the ICMD frontend using custom code. This enables customers to effortlessly interact with the chatbot using natural language, and for the chatbot to automatically populate the UI and set up simulations for the user.

The following figure (Figure 2) shows the second use case using visual conversation builder for Amazon Lex.

Figure 2: High-level architecture diagram of the solution described in use case #2.

Use Case #3: Generative AI Analyzer

The third use case is a generative AI analyzer that helps users identify unique data patterns in their simulation results. This tool is designed to provide deeper insights and facilitate more informed decision-making in the materials design process. By uncovering hidden trends and correlations, the generative AI analyzer empowers users to optimize materials properties and performance.

For the generative AI analyzer, AllCloud leveraged Amazon Bedrock and LangChain agents. Traditionally, LLMs have difficulty with mathematical and scientific reasoning. Using agents, AllCloud developed a tool for QuesTek to combine the power of natural language understanding by LLMs (such as Anthropic’s Claude 3 Opus) with the accurate and consistent code execution of agents on material’s science data.

The following figure (Figure 3) shows the third use case with Amazon Bedrock and LangChain agents.

Figure 3: High-level architecture diagram of the solution described in use case #3.

The integration of these AI-powered features has delivered significant benefits for QuesTek and its customers. It has streamlined research and development, reduced operational costs, and led to improved materials performance and innovation. This cloud-powered transformation of the ICMD platform has been a transformational for QuesTek’s business.

Results and Looking to the Future

With the financial support from AWS MAP, and the technical collaboration with AllCloud, QuesTek successfully launched ICMD for many corporate customers on the AWS cloud, helping them address their materials-centric R&D needs. Additionally, QuesTek is working towards offering an official solution on AWS GovCloud (US) to cater to United States government customers and their contractors.

The transition to AWS has provided QuesTek with numerous financial and operational benefits. By leveraging scalable infrastructure on AWS, QuesTek has achieved significant cost savings and improved efficiency. These benefits have enabled QuesTek to reinvest in further innovations and expand its capabilities using generative AI on AWS to better serve its customers.

Conclusion

Through its collaboration with AllCloud, and with support from AWS, QuesTek is not only enhancing its ICMD platform with cutting-edge generative AI capabilities, but also expanding its reach to a broader customer base. This strategic move positions QuesTek at the forefront of innovation in materials design and development, providing powerful tools and insights to their corporate customers.

Further Reading

• Amazon Bedrock User Guide
• Generative AI Innovation Center
• Amazon Bedrock to build and scale generative AI applications

.

.

AllCloud – AWS Partner Spotlight

AllCloud is an AWS Premier Tier Services and Generative AI Services Competency Partner that accelerates innovation and helps organizations fully unlock the value received from cloud technology.

Contact AllCloud | Partner Overview | AWS Marketplace