A-CX LLC

We used retrieval augmented generation to implement a custom LLM endpoint for answering internal questions, improving documentation lookup speed, and creating company-specific documentation and marketing posts.

Our client

Undisclosed

The challenge

Our client wanted to encourage their employees to explore AI. Staff wanted to use ChatGPT to create marketing materials, answer frequently asked questions, and provide investor summaries. They were especially interested in creating more targeted documents, using ChatGPT to quickly change the tone, target audience, or language of various articles, ultimately allowing them to produce a wider variety of marketing content and respond to customer inquiries faster.

How we helped

There were three primary goals for our solution:

1. Allow staff to use an LLM to create documents without requiring significant work gathering the information needed.
2. Ensure that documents and answers generated by the LLM do not include incorrect data.
3. Address any security concerns around data leakage.

Retrieval Augmented Generation

RAG is an approach to adding custom data to an LLM where instead of retraining the LLM on your custom data, you provide the data required to answer the query alongside the query itself. For example, if a user asks the LLM, “What does this widget do?” you might augment the query with information like “Widgets can be used for …” and then have the LLM answer the query using this provided information.

Analyzing User Queries

As a first step, we gathered a list of queries users commonly made. Our goal was to understand what types of data users typically requested and what sorts of questions they often asked. 

Determining Architecture For The Custom LLM

We determined that we could use Microsoft Azure to host a simple architecture composed of an Azure Static Web App, an Azure Function, and an Azure Cognitive Search database.

• Our static web app would provide an endpoint for users to enter queries, providing a site similar to the ChatGPT interface users were already familiar with.
• Our Azure Function would handle queries sent from the web app, and use our LLM to classify the queries to determine what data and query augmentation was needed.
• Our Azure Function would then query an Azure Cognitive Search database containing the client’s documentation to find the relevant documentation needed to answer the user query.
• The Azure Function would then send the augmented query and documentation to the LLM.
• Finally, the Azure Function would return the query result to the web app, where it would be displayed.

Indexing The Data

Our first step was to gather and index the client’s data into the Azure Cognitive Search resource. We performed this step manually, scraping information from their intranet and then uploading it to the cognitive search resource. We enabled semantic ranking, and selected indexes based on what information was most likely to be looked up in queries. 

Creating the Custom LLM Endpoint

Next, we created our Azure Function. We used Python to implement our function since it has libraries for querying both Azure Cognitive Search and OpenAI resources. It also had the nice advantage of allowing us to reuse code from our previous testing with our Machine Learning Studio notebooks.

To store our code and deploy our function, we used Azure DevOps, a git repository that smoothly integrates with Azure resources. This greatly simplified the process of creating development and production functions for testing, as well as ensuring that we had a continuous development environment to keep things up to date.

Adding A Web Interface

Finally, we created a web interface using React, having our design team work directly with the client to ensure consistency with the rest of the client’s in-house tools. We once again used Azure DevOps to store and deploy the web app. Its direct integration with Azure means we could quickly deploy updates and manage the web application via code.

Results

Our custom LLM solution was able to safely and securely help users answer a variety of internal questions, making it easier for our client to create summaries of important documents and answer customer queries. We were also able to reduce the workload for their sales and marketing teams significantly.

Languages

• Python
• Javascript

Frameworks

• OpenAI
• Azure Document Search
• React

Tools

• Visual Studio Code
• Azure ML Studio

Cloud

• Azure

We pioneered a VR training tool to transform heavy machine operator training, making it safer, more accessible, scalable, and cost-effective.

The challenge

Industries relying on heavy machinery are primed for a revolution in training. Traditional methods, although effective, have cost, logistics, safety, and scalability constraints. In collaboration with our client, we’ve crafted a transformative VR solution that dovetails with existing training programs to enhance their efficiency, safety, and accessibility.

How we helped

Our transformative project started with us meticulously crafting a high-definition, true-to-scale 3D model of an excavator.

This detailed model was then brought to life with interactivity applied in Unity. Through precision, every aspect of the virtual excavator mirrors its real-life counterpart, creating an authentic simulation that captures the experience of operating the machine.

Within the VR environment, trainees perform a safety inspection before operating. Once in the operator’s seat, they’re presented with a realistic view of the controls.

Reaching out, trainees can grasp and manipulate the joysticks, executing operations such as driving, turning, and controlling the boom and bucket, just as they would in real life.

Integrating real-world mechanics and physics into the VR platform delivers an immersive, practical, and engaging training experience.

Our breakthrough is this ingenious fusion of technology and practicality, which is driving the future of heavy machinery training.

Results

Our VR training tool is not just a revolution for heavy machinery training, it provides a safe and realistic environment for hands-on learning. It’s cost-effective, accessible, and ripe for industry-wide adoption.

Notably, it fosters safer training, minimizes accidents, and helps improve safety records. Our VR training tool marks a significant step towards a more efficient future in heavy machinery industries.

This case study is a testament to our commitment to product innovation and digital transformation in the heavy machinery industry.

Languages

• C#

Frameworks

• Unity
• Scrum Development

Tools

• Visual Studio
• Visual Studio Code
• Blender
• Figma
• Photoshop
• Bitbucket

Hardware

• Meta Quest 2

We implemented a TCP Proxy with decision-making sidecars to provide security when information/data crosses the boundaries of clouds/networks. The objective was to maintain low latency, design for scalability, and have no impact on the overall performance.

Our client

A leading global and highly regulated company headquartered in the USA.

The challenge

Our client managed critical and sensitive information in numerous on-premises and cloud applications and databases, lacking secure connectivity between clouds. Their challenge was that each application needed to secure communication between applications themselves. Security is paramount during communication with access control systems and must comply with industry standards. Our client needed a better, more secure solution and one less prone to human error.

How we helped

We worked directly with the client to develop a TCP proxy as a tunnel of encrypted data to secure communication between services deployed in the public and private cloud (on-premises). We had two teams working on the project, one to set up the environment, while the other developed within that environment. Our teams tackled the different aspects of development in the entire software development life cycle involving the product’s design, implementation, testing, and release.

Results

We implemented a tried and tested solution to a common problem of the exchange of information across boundaries, rather than each engineering team re-inventing the wheels.

Our solution was a highly available, secure, robust, and scalable system that increased the security of our client’s infrastructure by controlling and monitoring the communication between boundaries. It currently services the communication of thousands of applications making millions of requests between them. 

Languages

• Java
• Typescript

Frameworks

• Springboot
• Undertow
• Terraform

Cloud

• AWS
• GCP

IDEs

• IntelliJ IDEA

We secured communication between services while lowering our client’s operational maintenance cost by taking advantage of the Public Key Infrastructure (PKI) design and microservices.

Our client

A leading global, highly regulated service company headquartered in the USA.

Their challenge

Our client was moving their infrastructure from in-house to cloud-based services like AWS, GCP, and Azure. They needed a scalable solution to allow multiple services to communicate securely.

How we helped

We deployed a PKI to meet the robust security needs and to automate operational processes at scale. Our technical team was accountable for the implementation, testing, and production deployment of every component.

We then worked directly with our client’s technical team to integrate the microservices from the design, using the latest development, reliability, and DevOps standards.

Results

Our highly secure, automated, state-of-the-art framework supports millions of simultaneous Microservices. It also reduced our clients’ manual processes and human error margin, resulting in lower operational costs and increased efficiency.

Languages

• Java
• Javascript

Frameworks

• Springboot
• Webflux
• Terraform
• Multiple proprietary frameworks

Cloud

• AWS
• GCP 
• Azure

IDEs

• IntelliJ IDEA

Step into the future of forestry with our cross-platform enterprise fleet management solution. Enhance your current operations, predict future output, and extend your competitive advantage.

The challenge

Today’s forestry industry is worth trillions of dollars. Like many other industries, it relies on telemetry data from its fleet to monitor performance and efficiency, and make data-informed predictions. Most companies utilize a multitude of brands in the field like John Deere, Caterpillar, TimberPro, Komatsu, Hitachi, and Ponsse. Each specialist machine has the ability to transmit telemetry data to the cloud.

The challenge is that different brands of machines transmit differently formatted telemetry data, leading to a significant amount of time and resources to aggregate, normalize, and visualize the data into a meaningful representation that can provide an overview of the project. 

In today’s rapidly evolving business landscape, embracing intelligent automation is no longer a convenient option. It’s a necessity for staying relevant and competitive. By streamlining operations, your organization has the opportunity to accelerate growth and improve profitability, and sustain your competitive edge.

Our solution

The enterprise fleet management platform

We built a centralized fleet management platform that receives telemetry data from multiple brands of machines, normalizes it, and then presents the information as data visualizations. This allowed project managers, project coordinators, data analysts, etc., to access the data efficiently from any device (phone/tablet/desktop computer), freeing up the bulk of their time and resources to perform the actual analysis and planning.

Infographics

Data visualizations were the primary method for delivering information – by request. We held customer feedback sessions with the end-users to listen to their needs and requests and understand how these features would benefit them. In the end, our dashboard gave the user the ability to zero-in on one machine, a subset of machines, or view the entire fleet. Users can view performance over a day, week, month, or a custom date range. They can look at performance by region, or compare performance between multiple regions over the same date.

Real-time analysis

Project managers can now instantly assess the number of machines that need to be refueled, and order the truck to be on-site at the right time, minimizing downtime. Maintenance intervals can be scheduled well in advance, giving the PM the ability to reorganize tasks to accommodate.

Skills

• Primary & secondary research
• Workshop facilitation
• UX/UI design
• Prototyping
• Frontend development
• Backend development

Frameworks

• Design thinking
• Business model canvas
• Value prop canvas
• Lean startup
• Scrum

Technologies

• React Native
• Next JS
• Non-relational DB
• CSS, HTML

Tools

• Figma
• Miro
• Visual Studio Code
• IntelliJ IDEA
• Github

How we did it

At A-CX, we deliver value to our clients. Before taking on any project, we thoroughly analyze the business and its challenges. This is to ensure we can make a meaningful impact. We only ever engage if we are confident that we can meet our client’s specific needs and goals. 

Service Design

To provide the best possible service, our Service Design team gains a thorough understanding of the four crucial components of the organization: People, Processes, Products, and Partners. Our team engaged in workshops with representatives from various teams in the organization, actively listening and posing strategic questions. We also involved relevant stakeholders in these workshops. Sometimes, this even included representatives from the Ministry of Forests, who ensured that our early ideas were aligned with their specific requirements.

Design Sprints

After thoroughly understanding the organization through our Service Design process, we embarked on a collaborative innovation phase called Design Sprint. Our software engineers, designers, and project coordinators collaborated with members of our client’s organization to identify pressing challenges and rapidly ideate, prototype, and test the most promising solutions. This approach leveraged the expertise of our client’s industry and our experience in parallel industries, creating a shared ownership of the final solution and setting the path for ensuring success.

Creating an MVP

We presented the fleet management dashboard to our target users: project managers, coordinators, analysts, etc. Their feedback was integral in shaping the solution through iteration until it was ready for implementation as a pilot MVP. The goal of this pilot was to observe and measure the team’s performance with the new solution and compare the results to their performance before implementation to assess the solution’s impact.

Our client

A major consumer product company and household name (specific details obfuscated due to confidentiality)

The challenge

Early in 2021, our client noted that many professionals lacked the adequate tools to perform their jobs efficiently in the new work-from-home/hybrid model. They needed an unbiased perspective to find out more about what specifically was lacking from this audience’s tool set where they were experiencing friction in their workflows.

Gaining deep insights

Our client needed fact-based evidence that a sizeable market opportunity existed to garner support and mobilize their organization.

The Design Thinking framework helped us produce evidence-based insights and deeply understand the user’s challenge. We started with desk research, evaluating how current solutions on the market held up against the contextual shift in people’s working habits.

To explore this further, we recruited individuals who matched our client’s target persona from all over the USA to participate in remote 1-on-1 interviews where we learned about their experiences and what they were struggling with (and, in some cases, succeeding).

Our research insights and the corresponding data presented our client with the evidence to rally support within their organization and the confidence to proceed to the next phase: exploring solutions.

Skills

• Primary & Secondary Research
• Workshop Facilitation
• Industrial Design
• UX/UI Design
• Physical Prototyping

Frameworks

• Design Thinking
• Business Model Canvas
• Value Prop Canvas
• Lean Startup

Tools

• Figma
• Miro
• Zoom

Exploring solutions

Seeking the right solution is an exercise in divergent thinking; one must explore all the options, leave no stone unturned, and find unique angles. We took an effective and efficient approach by running a co-creation workshop to generate ideas. One must explore all the options, leave no stone unturned, and find unique angles. We took an effective and efficient approach by running a co-creation workshop to generate ideas. 

We invited a diverse team from our client’s multiple business units to help us find solutions from various perspectives. Each concept was measured against stringent criteria based on insights and the user’s jobs-to-be-done from the previous research phase.

When the workshop was over, our team shaped the top ideas into creative territories: loosely constructed concepts designed to stimulate discussion and further refinement. To help visualize these concepts, we produced low-fidelity prototypes and ran an additional round of interviews with participants matching our target profile to test our ideas. 

The feedback was insightful, helping us direct our efforts and prioritize features for the customers.

Validating our solution

Well into the convergent phase of the design thinking process, we aimed to test a solution and conclusively address the target audience’s challenges. 

By applying what we’d learned from the previous two rounds of interviews, we iterated on our concepts to produce higher fidelity prototypes, this time comprising hardware and software.

We then conducted in-person interviews at our Vancouver studio, asking participants to use our prototype and walk us through their experiences. The room was equipped with a camera array to document our participant’s feedback, body language, and facial expressions and record their actions in the software.

Our evaluation criteria of the potential solutions hinged on the following factors:

• Desirability (Do customers want and need this?)
• Feasibility (Can it be achieved with existing tech?)
• Viability (Can we monetize this?)

Results

Our team delivered conclusive, customer-validated evidence that our proposed solution addressed their customer’s needs, was feasible to produce and had a viable business model to justify their actions and make a return on investment.