Transforming AI Research into Scalable Solutions

When I first met Advai, the team consisted of pioneering machine learning researchers eager to bring their innovations to market. My goal was to assist in this transition by helping them to develop a compelling demo to attract businesses, secure initial customers, and drive subscriptions to their service.

My research goals and approach

My research objectives were threefold:

  1. Developing Personas: I aimed to help the team understand the goals, needs, and challenges of their end users. By adopting a persona-based perspective, the development team could better cater to different user requirements.
  2. Creating a User Journey Map: I mapped out an initial walkthrough of the platform to identify the necessary user interface elements that would support user interactions effectively. This approach helped us refine the demo to address the most critical needs of the end users.
  3. Designing User Interface Elements: I produced sketches, patterns, and wireframes to facilitate meaningful discussions and guide the development process.

I employed a mix of qualitative research methods, actively participating in discussions with customers to understand their needs and experiences. Additionally, I observed and conversed with the team to grasp their processes, aiming to understand team dynamics and workflows in context.

My initial insights and persona

Advai’s current process involved conducting extensive research to identify flaws in customers’ AI models and presenting the findings in a detailed document. However, this document required significant expertise to interpret, often conflating business and technical decisions. There was a clear need for a more effective way to communicate complex ideas to both technical and non-technical users.

To address this, I created two personas:

  1. Non-Technical End User: Seeks a competitive advantage through AI insights.
  2. Technical End User: Focuses on selecting the best AI model for their needs.

For a quick win, I tailored the document to appeal to both personas. I structured it with clear sections to emphasise key points and included additional details for technical readers as needed, ensuring accessibility and actionability for all users. Moreover, I ensured that the document’s style reflected Advai’s brand values of creativity, intelligence, and technological innovation.

Design principles to guide the technical team

To guide platform developers and prevent scope creep, I distilled the user research into three key principles:

  1. Automation: Focus on automating tasks to free up specialists to engage in the complex, rewarding work they enjoy, and eliminate mundane tasks like report writing.
  2. Lifecycle management: Ensure every feature supports decisions about whether a model adds value to innovation, pre-deployment and post-deployment processes.
  3. Simplify and Clarify: Present results in straightforward, real-world terms, making complex information accessible and understandable for all users.

Essential user requirements

I led and supported various creative sessions with the development team to define the platform’s essential features. The platform needed to support the following user tasks and needs:

  • Scheduler: Manage and automate tasks efficiently.
  • Complex ML Systems: Handle intricate machine learning systems.
  • Scalability: Ensure the platform can scale to meet growing demands.
  • Model Comparison: Allow users to compare different AI models effectively.
  • Pre and Post Deployment Metrics: Provide metrics for both pre-deployment and post-deployment stages.
  • User-Defined Metrics: Enable users to define and track their own metrics.
  • Transfer to High Side: Facilitate secure data transfer to high-security environments.
  • Hardware-Related Advice: Offer guidance on hardware requirements and optimisations.
  • Knowledge Transfer: Ensure effective transfer of knowledge within the user community to benefit individuals, teams or organisations.
  • End User Feedback: Collect and incorporate feedback from end users to continuously improve the platform.

Architecting a solution by understanding how people will use it

Next, I mapped the various stages of the user journey, defining and refining the user interface and user experience for each step by iteratively testing with technical and non-technical end users. I identified opportunities to enhance each stage and incorporated suggestions to tailor the experience for different user types, including power users. For example:

  • Onboarding and initial configuration wizards tailored to different user profiles for a seamless introduction to the platform
  • A dashboard to support technical and executive decision-makers by providing key metrics and outputs about the health and status of their AI suite.
  • Task builders to simplify the process of building and running tasks and detailed technical reports for data scientists and machine learning engineers
  • A community platform for knowledge sharing and joining projects within the community.

Clarifying communication

I conducted a “When I Say, We Mean” session to clarify terminology around robustness, resilience, and reliability. Additionally, I led a workshop focused on automation, working with the team to define sector-specific implications (“so what’s”) for adversarial attacks and sector specific risks and considerations.

Demo attracts first customers

In collaboration with the tech lead, I defined wireframes that underwent review, approval and then were passed to a user interface designer/developer for refinement. Together, we shaped these wireframes into a functional demo aimed at attracting customers. 

This demo highlighted two key products: Advai Insights, where AI models are presented within a risk and compliance framework, monitoring against business objectives, global standards, and risk appetites; and Advai Versus, which automates tasks such as stress-testing, red-teaming, and evaluating AI systems for critical failure to provide assurance and improvement. 

This demonstration showcased Advai’s expertise, craftsmanship, machine learning proficiency and emphasis on data-driven outcomes. Importantly, it served as a pivotal tool in attracting the company’s first customers.

Scale and grow

Advai now offers a comprehensive suite of tools designed to test, evaluate, and build trust in AI systems, all grounded in bleeding-edge research. Their tagline captures their unique approach: “We don’t make AI, we break it.”

Check out their unique service here:

Leave a Reply

Your email address will not be published. Required fields are marked *