Richard Walls

Challenge: New users found the onboarding process overwhelming and unclear where to begin.

My Role: I designed a clear, prominent "Add to dashboard" call to action as the first step, paired with a concise explanation of its purpose. This provided a single, obvious starting point and mapped the user journey from the very first interaction.

Impact: This approach accelerated platform adoption and reduced friction, making it easy for users to start using the platform confidently.

Challenge: Users needed to select from many deployment spaces, risking information overload.

My Role: I applied progressive disclosure, presenting only necessary information at each step. Locations were organized with clear labels like "Name" and "Environment type" to support easy identification and selection.

Impact: This guided, step-by-step approach reduced cognitive load and made onboarding more approachable.

Challenge: Users needed clear guidance to select the correct deployed model from a list of assets.

My Role: I designed the screen to display all relevant details—"Name," "Type," "Asset," and "Last modified"—so users could make informed choices quickly and confidently.

Impact: This clarity in information architecture enabled efficient selection and contributed to a simplified onboarding journey.

Challenge: Users needed to understand technical storage options and make the right choice for their workflow.

My Role: I provided plain-language explanations for each storage type and added contextual help for environment types, making technical decisions accessible and understandable.

Impact: This approach empowered users to make informed decisions and reduced confusion during onboarding.

Challenge: Users needed a way to verify all their setup choices before finalizing onboarding.

My Role: I designed a comprehensive review screen summarizing all selections, with a clear "Edit" option for corrections.

Impact: This step prevented errors, increased user confidence, and contributed to a reduction in support tickets related to setup confusion.

Challenge: AI developers needed a way to efficiently evaluate and compare multiple prompt templates for different tasks. The process was previously complex and overwhelming for new users.

My Role: I designed the step-by-step evaluation flow, breaking down the process into clear stages—Setup, Prompt Templates, Metrics, Test Data, and Review. I applied progressive disclosure principles to guide users through each step, reducing cognitive load and making the workflow approachable for both new and experienced users.

Impact: The new interface improved onboarding, increased user confidence, and accelerated platform adoption by making complex tasks more intuitive and manageable.

Challenge: Users often selected prompt templates with mismatched variables, leading to failed evaluations and frustration.

My Role: I implemented real-time error prevention with clear, actionable feedback. When a variable mismatch is detected, the system immediately displays a prominent error message with microcopy explaining the issue and guiding users on how to fix it.

Impact: This proactive approach prevented invalid configurations, saved users time, and reduced frustration by providing clear guidance directly within the UI.

Challenge: Technical AI evaluation metrics like BLEU, ROUGE, and F1 scores were confusing to both technical and non-technical users.

My Role: I provided plain-language explanations directly within the UI, making complex concepts accessible. For example, the description for "Generative AI quality" clarified how metrics work without requiring technical jargon.

Impact: This approach broadened the platform's appeal, reduced support requests, and made the evaluation process more inclusive for all users.

Challenge: Users struggled to prepare and map their data for evaluation, leading to errors and confusion.

My Role: I designed a streamlined "Test data" screen with clear guidance on required columns and intuitive dropdowns for input and output mapping, simplifying the technical process.

Impact: This improved information architecture ensured users could easily configure their test data, leading to more accurate AI asset comparisons.

Challenge: Users often modified test data after evaluations, leading to confusion about outdated results.

My Role: I implemented dynamic feedback, providing timely warnings when test data changes, informing users that evaluations would need to be re-run to reflect new data.

Impact: This proactive guidance prevented confusion, ensured users always worked with current evaluation scores, and improved overall user confidence.

Challenge: Users needed a clear and intuitive way to compare performance metrics across different prompt templates, which was previously complex and time-consuming.

My Role: I designed a comparison screen that utilizes both bar charts and a detailed table to display performance metrics. This multi-modal presentation allows users to easily identify which AI assets perform best according to each metric.

Impact: The design accelerates development cycles and reduces user friction by providing actionable insights in a transparent and intuitive format, enabling users to make informed decisions quickly.

Challenge: AI developers needed a streamlined way to evaluate prompts with labeled data, but the process was complex and overwhelming for users unfamiliar with evaluation workflows.

My Role: I designed a structured, step-by-step user journey that breaks the evaluation process into clear, manageable phases: "Set up," "Configure evaluations," and "Review and evaluate." I applied progressive disclosure principles to present the overall process in digestible chunks, allowing users to focus on one aspect at a time without cognitive overload.

Impact: The systematic approach reduced user confusion and enabled both technical and non-technical users to successfully navigate complex AI evaluation tasks, improving platform usability and adoption.

Challenge: Users needed to configure advanced evaluation metrics like ROUGE with complex parameters, but presenting all options upfront would overwhelm users and hinder task completion.

My Role: I designed a progressive disclosure system that reveals detailed configuration options (like "Use aggregator," "Use stemmer," and threshold settings) only when users actively select specific metrics. I created contextual help through information icons that provide immediate explanations for complex terms, making advanced settings accessible to non-technical users.

Impact: Users could systematically configure complex evaluation metrics without feeling overwhelmed, enabling more precise prompt assessments while maintaining an intuitive user experience.

Challenge: The platform lacked any way for users to visualize and interpret AI evaluation results. Users received raw data outputs but had no visual interface to understand their model's performance patterns or make data-driven decisions about prompt optimization.

My Role: I designed the first comprehensive results visualization feature for the platform, creating an interface that combines high-level summary metrics (like the 0.96 Mean faithfulness score) with detailed visual distributions through histograms. I built a new layered information architecture that shows both aggregate performance and individual transaction details, implementing progressive disclosure so users could start with overview insights and drill down into specific instances using the new "Analyze" action for deeper investigation.

Impact: The new visualization feature gave users their first ability to interpret evaluation data visually, transforming what was previously inaccessible raw output into actionable insights and enabling confident AI model assessment where none existed before.

Challenge: Users needed a systematic way to create prompt templates that would yield consistent, evaluable outputs, but lacked guidance on prompt engineering best practices within the platform.

My Role: I designed the "Prompt Lab" interface with clearly defined sections for "Set up" (instructions), "Examples" (few-shot learning), and "Try" (immediate testing). I integrated contextual help through information icons and seamlessly connected prompt creation to evaluation workflows by embedding "Evaluate" and "Save work" functions directly in the interface.

Impact: The structured environment enabled users to build effective, evaluation-ready prompts while learning best practices, resulting in higher-quality templates and more meaningful evaluation results.