Driving Business Value with AWS Generative AI

August 17, 2023

For over 20 years, Adastra has worked in the data and analytics space, developing custom artificial intelligence (AI)-based solutions for clients across different sectors and applications. In cooperation with AWS and their array of innovative AI and ML solutions, Adastra leverages new advances in technology to accelerate processes. By moving quickly from evaluation to implementation, Adastra delivers innovative and practical generative AI-based solutions for clients and their customers.

AI: Generative vs. Predictive

Generative AI is a hot topic: we’ve been told it will change the way things are done and that it will play an increasingly prominent role in the tools we use to work and live our daily lives. But how can we apply the fantastic power of artificial intelligence (AI) and machine learning (ML) to improve business processes?

Generative AI’s superpower is its ability to create new content. By analyzing patterns and features from historical data, it can be trained to create new content such as images, audio and text based on that harvested knowledge.

Before generative AI was widely available, data scientists typically focused on AI’s predictive capabilities. While generative AI focuses on creating something new based on the past, predictive AI makes forecasts or predictions by training on historical data to predict outcomes.

Generative and predictive AI use different approaches and algorithms and, depending on the use case, one or both methods can be integrated into a solution to improve and automate systems and processes.

Adastra’s AI Solutions

Architecting and implementing data, analytics and AWS Cloud-based solutions is at the heart of our business. Adastra offers a range of generative AI-based solutions that take full advantage of the power of AI and ML technologies offered by AWS. In addition to bespoke, designed solutions, we have developed three repeatable and highly adaptable offerings that can be quickly implemented to leverage the power of generative artificial intelligence on AWS.

Smart Chatbot Technology

Traditional chatbots lack context and natural language processing abilities, leading to low success rates.

Adastra’s Intelligent Chatbot Platform uses Amazon Lex to improve successful bot implementation and outcomes, leading to better user experiences, and less manual intervention from live agents.

Our intelligent chatbot can summarize content, generate conversational analytics and enable conversational interfaces in applications, understanding intent and context and automating simple tasks across multiple languages.

Intelligent Search Platform

Accurate, data-driven insights are vital for organizations in today’s competitive landscape, but traditional search mechanisms often fall short in extracting them, as they require precise criteria, lack context, and overlook relationships between data sources. Adastra’s Intelligent Search Platform combines traditional search algorithm outcomes with generative AI techniques to allow organizations to extract valuable business insights from both structured and unstructured data sources while improving search relevance within internal document libraries. The platform can also provide summarized responses to user questions in an easily consumable form.

Intelligent Content Creator

Creating collateral for new product launches and specific customer groups can require significant manual effort, and due to limited time, organizations are often forced to make do with outdated or generic content.

Adastra’s Intelligent Content Creator uses the power of generative AI and AWS to craft timely sales and marketing collateral, ensuring customers are informed and engaged.

Adastra customizes all our solutions to our customer’s unique business needs, using the tools that our clients are already using or are familiar with. We also provide managed services in the form of ongoing support and maintenance for each solution.

“When we implement a solution, we can surface generated insights and outcomes within the tools our clients are already using,” points out Katya Dunets, AWS Lead Sales Engineer, Adastra.

“For example, if they’re using Salesforce, we’ll integrate our sales and marketing content generation outcomes with their Salesforce platform.”

AWS Generative AI Use Cases

Adastra has been implementing successful solutions leveraging AI, ML, and generative AI, long before their recent popularity. In the past few years, we have implemented numerous innovative customer-driven solutions.

Some examples include:

• Building a generative AI-driven chatbot service for a major national organization that enabled users to search for content across all of their internal documentation, including PDF and image-based documents. Before the implementation, users had to sift through historical documents to manually generate new templates and content.

Our generative AI solution was accessible to users through a chatbot interface and enabled intelligent search with easily consumable search summarizations/responses through generative AI. It then used that information to create templates for new documents based on similar topics drawn from a database of older documents and pre-populate them with new content.

• Developing a sales and marketing content generation system for a major Canadian beverage producer that allowed them to track specific products and then target those products for additional marketing activities.

The technology applied AI to generate sales transcripts in natural language that automatically highlighted the company’s top products, using stock-keeping unit (SKU) codes, that the company wanted to promote to retailers.

The technology allowed the company to shift its marketing focus towards promoting new or slower-moving products amongst their sales networks.

• Looking to evaluate the net promoter scores for industry events, we worked with an international pharmaceutical company to develop a system that could quickly and efficiently process the feedback from events they were hosting.

Automated evaluation of the marketing metrics gave the company an analytical tool that enabled them to better tailor future events for their audience.

High-Value Domains

Generative AI can be used in many ways, but below are some areas that help organizations see the highest value.

Content Creation and Management

Adastra’s customers must often deal with large and unwieldy corpora of unstructured data and seek the best way to sift through, analyze and organize that valuable information.

We also help clients develop content creation strategies by applying generative AI to improve written communication and SEO best practices, as well as brainstorming. We can help writers organize their content plans and develop editorial apps that identify areas for improvement and suggest the best time to publish content.

Workforce Training

Generative AI can create personalized learning experiences for individual employees based on their specific needs and learning styles. It can also provide content, such as training videos, documents, and interactive exercises, tailored to the employee’s level of expertise.

Efficiency Improvement

Generative AI offers many features to speed up organizational processes, including automating repetitive tasks and providing instant responses to common questions.

This applies to finding efficiencies for code developers by automating code documentation and generating templates for new code.

By helping automate processes such as code reviews or building new code bases, generative AI helps accelerate development efforts when starting new projects. It can also dramatically improve the speed and quality of code documentation and QA testing. This can have significant positive impacts on developer effciency.

Our ‘Go-To’ Generative AWS AI

Adastra leverages a variety of AWS tools to deliver effcient artificial intelligence solutions for our clients. Some of our favorites include:

Amazon Trianium

Provides accelerators for training deep learning models in the cloud, reducing costs by 50%. The platform is scalable and can be used across a broad set of applications.

Amazon Bedrock

Provides access and private customization to foundational models (FMs) from Amazon and various AI startups, to build and scale generative AI applications. Foundational models are accessed through serverless APIs, ensuring the security and privacy of critical organizational data.

Amazon SageMaker JumpStart

An essential tool at Adastra for any data science, model training and development jobs, and most generative and predictive AI projects. The platform is fully managed and designed to accelerate the process of building and deploying machine learning models with pre-built AWS solutions and code samples for a wide range of use cases.

Client Data and Security

Privacy and security are top-of-mind concerns inside and outside of generative AI projects.

We start all projects with a thorough assessment and work closely with our clients to identify the specific data that will be used for a project. Once identified, we establish clear guidelines for how that data will be used and protected. This includes enabling secure data storage on the AWS Cloud and local systems via access controls and data encryption. And as part of our managed service agreement, we also conduct regular security audits and vulnerability assessments to identify potential or existing threats.

The processes built into all our projects, including the ones that implement generative AI technology, operate on a similar methodology regarding data security and privacy. Our clients rely on our abilities to secure personally identifiable information (PII) or customer-specific data.

We are extremely diligent in this area and work carefully to protect PII. We don’t want our AI models to be trained using sensitive customer information. To that end, we ensure all vestiges of PII are removed before training our models.

We also offer our clients training and support so that they have a thorough understanding of what is required of them to ensure that their data and their client’s data remain private and secure. This is a critical reason for ensuring that they have the training and documentation they need. It’s also essential to ensure that clients can easily access the new implementations without having to manage code bases or retrain AI models outside of pre-determined intervals.

Lastly, all of our generative AI solutions are deployed using private instances of models available through AWS, ensuring your organization’s critical data stays protected from data breaches and public consumption.

AI and Bias

The issue of bias when implementing AI technology is linked primarily to data quality. If there is an inherent bias in the underlying data you use to train your algorithm, that bias will show in your model.

That is why we prioritize using diverse and inclusive data sets when working with clients. On any project that implements or leverages generative AI technology, we spend considerable time on exploratory data analysis, looking at trends and skews within our data before we use the data for model training.

Adastra has a dedicated ML operations team that sets up processes to monitor changes and trends over time from incoming data. For example, if models are trained monthly, we will analyze the data at each training iteration to ensure that no new biases are introduced into the updated learning models.

Generative AI – The Challenges

As a company with a long history in the AI and data space, the main challenge we encounter is the lack of sufficient high-quality data to train AI models adequately. In these instances, one of our main tasks is to help our clients identify and acquire the necessary data. We then offer guidance on how to collect it and how much data is needed to properly train and generate an accurate AI system.

Another priority is ensuring that the available data is clean and well-processed before being used by the ML system. This involves the pre-processing steps like evaluating for biases that help to ensure that results will be accurate and valuable as the client moves forward.

The issue of data quantity can be addressed by using generative AI techniques to produce new data sets. But training using this method is computationally intensive and requires large amounts of high-capacity computing power. That is also why AWS scalable solutions are so helpful, providing computing power on demand so that clients only pay for what they use. For many Adastra clients, model explainability is also a critical concern.

They want to know how and why the AI generated a particular prediction or outcome. However, using deep AI learning methods may not always be the best approach to understand what feature contributed to a particular prediction. The rationale used by a complex AI to generate a specific decision or output may become too diffcult to pinpoint.

As a result, we need to have a detailed conversation with our customers at the start of the development process about the level of ‘explain-ability’ they require. Sometimes that may mean choosing algorithms and techniques that might sacrifice performance and lower complexity to make the AI’s logic more apparent and explainable to the end users. Typically, in cases where explainability is prioritized, we’ll choose algorithms that allow for a more straightforward logic and leverage features such as visualizations so that the reasoning used to generate a result or outcome is more transparent.

Leveraging Ethical AI

To mitigate concerns about the bias and explainability of artificial intelligence, Adastra leverages an ethical AI framework, using Human in the Loop (HIL) to create, validate and update AI models.

Retraining models with customer data is also an essential step and we often recommend that clients start smaller and create a minimum viable product (MVP) before moving into a potentially large and complex AI solution. This allows clients to prove a return on investment and get the all-important buy-in from stakeholders for future implementation.

As generative AI functionality becomes increasingly embedded in more business functions, it will generate new possibilities and capabilities. At Adastra, our goal is to be at the forefront of this evolution, deploying the powerful and cost-effective services and products offered by AWS to speed up implementation and deliver meaningful tailored solutions for our clients.

About Adastra

Adastra helps organizations optimize business intelligence and leverage AI-driven solutions to gain value from business data. As an AWS Advanced Tier Services Partner with over twenty years of experience in the data and AI space, we provide trusted insights that help businesses and IT leaders make better decisions every day.