AI, Product Design & Real‑World Deployment
This page brings together the projects where I’ve used AI to turn ideas into real products, from spotting gaps where no good solution existed, to researching, designing, and deploying tools that people can actually use. Each project reflects my approach to building with purpose: combining AI with product thinking, accessibility, and practical delivery. Here you’ll find the stories behind PBIX A11y, Galaxy Maths Quest, and the new ideas I’ll continue exploring as I grow this space.
1. PBIX Ally
Context
As a Power BI accessibility advocate, I kept seeing the same problem: report authors wanted to build inclusive dashboards, but the tools available were fragmented, manual, or incomplete. Accessibility guidance existed, but it lived across blogs, PDFs, and scattered documentation, and none of it helped people check their PBIX files in a practical, automated way.
With my background in digital accessibility and data visualisation I knew exactly what needed to exist. And with Copilot prompting and Lovable’s build environment, I finally had the means to build it.
Problem
Real‑world usage revealed persistent barriers:
• No automated accessibility auditor for PBIX files
• Manual checks were slow, inconsistent, and easy to forget
• Text contrast issues were common
• Non‑text contrast was rarely checked
• Alt text was often missing or unclear
• Tab order was frequently illogical
• Decorative shapes were being read by screen readers
• Font sizes not adequate for canvas size
• Corporate themes introduced contrast and scaling issues
• No GDPR‑safe solution existed, everything required connectors
Even experienced authors struggled to meet WCAG AA, and organisations had no scalable way to validate reports.
Product Strategy & Approach
I approached PBIX A11y as a product, not just a technical solution.
I defined a clear product vision: to create a practical, automated, and privacy‑safe accessibility platform tailored specifically to Power BI. The goal was not just to check compliance, but to provide meaningful, actionable insights that users could trust and apply in real‑world reporting scenarios.
To achieve this, I translated accessibility challenges and WCAG 2.1 AA requirements into a structured product framework and prioritised backlog, focusing on features that would deliver the highest user value while remaining technically reliable within Power BI’s constraints.
I used Copilot as my prompting partner and Lovable as the build environment to accelerate development, but the foundation of the product came from something far more specific: my deep knowledge of WCAG and a detailed understanding of what accessibility checks are actually possible inside a Power BI report.
Before writing a single prompt, I mapped every WCAG 2.1 AA requirement against Power BI’s technical reality, what metadata exists within a PBIX file, what can be reliably tested, what must be inferred, and what cannot be detected programmatically. I also defined which checks would meaningfully improve accessibility without introducing false positives.
This step was critical. Power BI is not a web page, and WCAG cannot be applied blindly. I had to define, rule by rule, which checks were feasible, how they should function, and how results should be interpreted to ensure outputs were accurate, credible, and actionable.
With this foundation in place, I prioritised delivery of a focused Minimum Viable Product (MVP) to validate the core concept and deliver immediate value. The MVP centred on the most impactful and reliable accessibility checks:
- Text and non-text contrast
- Alt text completeness
- Titles (Page, Visual and axis)
- Decorative shape detection
- Logical tab order
- Clutter
- Font scaling
Most importantly, all these checks had to respect data privacy.
Iteration & Product Growth
Once the initial Report Auditor MVP was stable, I adopted an iterative, feedback‑driven product approach, using real user insight to refine and expand the platform.
PBIX A11y evolved into a broader accessibility ecosystem through the addition of standalone tools, each prioritised based on user needs and aligned with the same WCAG‑based, Power BI‑specific logic:
- Colour‑blindness Simulator
(including image upload and HEX‑based simulation) - Alt Text Generator
(static WCAG alt text + dynamic DAX alt text) - Colour Contrast Checker
(Relative Luminance + APCA methods) - Theme Accessibility Audit
(contrast, font scaling, chart titles, axis titles, UI colours) - Cognitive Load Score
(layout density and visual complexity) - Minimum Font Size Calculator
(a method using proportionality to the canvas size) - UI Layout Calculator
(sizes for canvas margins, buttons, padding, shapes, icons and border radius) - Visual Design Audit
(combined visual, structural, and readability checks)
Each feature was shaped and refined through feedback from Power BI developers, ensuring the product remained practical, predictable, and trustworthy in real‑world use.
I also ensured the product remained aligned with platform evolution. Following community feedback and Microsoft’s shift toward project‑based report structures, I prioritised and delivered support for PBIX, PBIR, and PBIP formats, keeping the product relevant and future‑ready.
While early feedback validated usability and functionality, further validation with users who rely on assistive technologies remains an important next step in the product’s evolution.
Throughout the lifecycle, I maintained clear product principles:
• WCAG 2.1 AA alignment
• GDPR‑safe behaviour (no uploads, no storage)
• Clear, actionable explanations
• Predictable UI
• Minimal visual clutter
• Accessibility first, always
Result
PBIX A11y is now a free, browser‑based accessibility platform used by Power BI creators, teams, and organisations to validate and improve report accessibility at scale.
It enables users to:
- Run automated accessibility audits on reports and themes
- Generate accessible alt text
- Simulate colour‑blindness
- Assess cognitive load and layout quality
- Apply WCAG guidance in a practical, actionable way
Following launch, I was invited to contribute to the Microsoft Community Blog, sharing the thinking behind PBIX A11y and driving engagement within the Power BI community. This created a strong feedback loop that directly informed product improvements.
. Within just two weeks of launch, PBIX A11y was nominated for the Inclusive Awards 2026 in the D&I Tech Initiative category in the UK, a recognition that validated both the need for the solution and the effectiveness of its design.
Wrap-Up
PBIX A11y began as a clearly defined gap in the ecosystem and evolved into a fully realised product through structured discovery, MVP delivery, and continuous iteration.
It demonstrates how strong product ownership, domain expertise, and the effective use of AI tools like Copilot can deliver a scalable, real‑world solution that improves how accessibility is embedded into data products.
2. Galaxy Maths Quest
Context
My son is neurodivergent and struggles to engage with content that doesn’t align with his interests or cognitive needs. The maths tools recommended by his school were visually cluttered, generic, and not designed for neurodiverse learners.
With a background in digital accessibility and data visualisation, I could clearly see why these tools were failing. They weren’t built with cognitive accessibility, engagement, or real user needs in mind.
This created an opportunity to design a more effective, inclusive learning experience from the ground up.
Problem
Homework sessions revealed several barriers:
- Too many distractions on each page
- Small fonts with poor character differentiation
- Weak visual hierarchy
- Overcrowded layouts
- No consideration for cognitive load
- No alignment with his interests
As a result, engagement was low, and learning outcomes were limited. The available tools were not designed for children with different cognitive needs.
Product Strategy & Approach
I approached Galaxy Maths Quest as a product, not just a personal project.
I defined a clear product vision: to create a calm, engaging, and accessible maths platform specifically designed for neurodiverse learners, where usability, focus, and motivation were treated as core product outcomes.
Using Copilot as a prompting partner and Lovable as the build environment, I accelerated development while maintaining full ownership of product decisions, design direction, and user experience.
I translated observed challenges into a structured product framework and prioritised backlog, focusing on reducing cognitive load, improving clarity, and aligning the experience with user motivation.
The initial MVP focused on Times Tables and was designed around key product principles:
- A galaxy theme aligned with user hyperfocus and motivation
- No ads, chat, or external contact (distraction‑free environment)
- WCAG 2.1 AA compliance
- GDPR‑aligned behaviour
- Clear typography and strong visual hierarchyPredictable layouts and interactions
- Minimal visual clutter
- Dark/light mode for user comfort
This MVP validated that engagement improved when accessibility, structure, and personalisation were considered together.
Iteration & Product Growth
Following MVP validation, I adopted an iterative, user‑led approach, expanding the platform based on real learning needs and observed gaps.
The product evolved into a series of structured “quests,” each targeting specific maths skills:
- Number Bonding
- Fractions
- Data Challenges
Each new feature was prioritised based on learning impact, usability, and cognitive accessibility, ensuring the platform remained focused and coherent.
I later expanded accessibility and reach by introducing full localisation in English, Portuguese, Spanish, and German, making the product usable across a wider audience.
Testing included real users, with feedback used to identify usability issues and refine the experience. I used Copilot to explore and validate potential solutions, iterating quickly while maintaining control over final decisions and quality.
Throughout development, I maintained clear product principles:
- Continuous improvement based on user feedback
- Accessibility and cognitive load reduction
- Predictable and consistent user experience
- Engagement through meaningful design (not distraction)
- Privacy and child‑safe interactions
Result & Impact
Galaxy Maths Quest is now a free, child‑safe, multilingual maths platform designed specifically for neurodiverse learners.
It provides:
- A calm, distraction‑free learning environment
- Accessible, structured maths challenges
- Narrative‑driven engagement through themed “quests”
- Support for multiple languages
Most importantly, it successfully supports sustained engagement and learning for its target users—including my son.
What began as a personal need has evolved into a scalable product with the potential to support many more children through thoughtful design, accessibility‑first principles, and responsible use of AI tools.
Wrap-Up
Galaxy Maths Quest demonstrates how product thinking, accessibility expertise, and user‑centred design can transform a common problem into an effective, real‑world solution.
It highlights the value of:
- Defining clear user problems
- Delivering focused MVPs
- Iterating based on real feedback
- Designing for inclusion from the outset
and shows how AI tools like Copilot can support, but not replace, strong product ownership and decision‑making.
Hello,
I’m Juliana
The creator behind Smart Frames UI, a space where storytelling meets strategy and dashboards do far more than display numbers. For over 15 years, I’ve been driven by a passion for uncovering the stories hidden within data and transforming complexity into clarity. Specialising in analytics engineering and data design, I focus on shaping well‑structured models, intuitive data flows, and visual experiences that help people understand information quickly and confidently. Through Power BI, I build insights that are not only visually compelling but also genuinely accessible, following WCAG‑aligned practices to remove barriers and ensure informed decision‑making is inclusive for everyone.
Let’s connect
Join the fun!
Stay updated with our latest tutorials and ideas by joining our newsletter, and receive a free Power BI accessibility checklist when subscribe.
Buy Me a Coffee to Fuel My Projects
3. STEM Demystified
Context
Throughout my work in technology, accessibility, and STEM outreach, I repeatedly encountered the same challenge: many young people, particularly girls, struggle to see themselves in STEM careers because they lack relatable role models.
Career resources often focus on job titles and qualifications, but rarely explore the personal journeys behind them. As a result, STEM can appear distant, intimidating, or only accessible to a narrow group of people.
I wanted to create something that would make STEM careers feel more visible, relatable, and achievable.
Problem
Several barriers consistently emerged:
- Limited visibility of women working across STEM disciplines
- Persistent gender stereotypes surrounding technical careers
- Lack of accessible career stories aimed at younger audiences
- Few opportunities for students to hear directly from diverse role models
- Career information often focused on pathways rather than lived experiences
- Underrepresentation of emerging and interdisciplinary STEM careers
These challenges contribute to a perception that STEM is “for other people” rather than a realistic option for everyone.
Product Strategy & Approach
I defined a clear product vision: a platform where any child, anywhere in the world, could find a woman in STEM who looked like them, came from where they came from, and had built a life in science, technology, engineering, or mathematics, at no cost, with no account required, and with their safety and privacy built in at every layer.
Using Lovable as the build environment and Claude as my thinking partner for architecture decisions, privacy design, and accessibility auditing, I was able to move at a pace that would not have been possible building alone. But every product decision, what to build, how data would be handled, who would have access to what, and what children would and would not see, was mine.
Before writing a single prompt, I mapped the full user landscape:
- Young explorers: browsing freely, no account, no data collected
- Teachers and educators: needing access to ambassadors and Q&A, with appropriate safeguards
- STEM Ambassadors: professionals sharing their stories and connecting with schools
- Organisations and charities: seeking visibility in a curated, approved directory
- Administrators: moderating content, approving applications, managing the platform
This mapping directly shaped the architecture. A role-based system with Row-Level Security as the primary enforcement layer, SECURITY DEFINER functions for sensitive data access, and twelve non-negotiable privacy guardrails embedded before any feature was built.
The MVP focused on the core experience: a 3D interactive globe showing role models’ locations, their individual story pages, and a public ambassador showcase, all publicly accessible without an account, built to WCAG 2.1 AA, and deployable on the live domain stemdemystified.com.
Key MVP principles:
- Privacy-preserving analytics with no third-party trackers, no cookies, no consent banner
- All role model content free and public: no account, no paywall
- No personal data collected from children under any circumstances
- UK GDPR compliance as a foundation, not a finish line
- Multilingual from the outset (English, Spanish, Brazilian Portuguese, German)
- The STEM acronym localised by language (CTIM in Spanish, CTEM in Brazilian Portuguese, MINT in German)
Iteration & Product Growth
Once the core public experience was stable, I expanded the platform iteratively, adding layers that served each user type without compromising the others.
The 3D globe evolved from a basic pin display into a full clustered, two-category system: red pins for newsletter role models, blue pins for historical women who changed STEM history. Each pin opens a popup with the woman’s name, discipline, country of origin, and a link to her story. Clustering prevents dense regions from becoming untappable. A reduced-motion fallback replaces the 3D globe with a static SVG world map for users who have motion sensitivity enabled.
The authenticated layer introduced four distinct account types, each with purpose-built dashboards:
- STEM Ambassadors: approved via an admin-reviewed application, with granular opt-in controls for public visibility, school directory listing, and virtual talk availability. Safeguarding declarations visible to school accounts only, routed through a SECURITY DEFINER function that returns booleans only, no raw check details.
- Schools and Educators: self-serve sign-up with immediate access to the ambassador directory, Ask a STEM Professional Q&A, and an aggregate-only student engagement dashboard. Individual student activity is never visible to teachers.
- Organisations and Non-Profits: directory listings with an admin approval workflow, a dedicated charities page featuring Technology Books for Children as the platform’s chosen charity, and a non-profit registration flow with explicit consent and GDPR-compliant data handling.
- Admins: a full moderation dashboard with content approval, ambassador assignment, notification dispatch, and audit logging.
Multi-profile support was added to allow existing users to hold multiple roles. An ambassador who also runs a non-profit can manage both from a single account with a profile switcher, without creating a new login.
Security hardening by addressing all the warnings flagged by Lovable, with support from Claude. This included tightening file‑upload permissions, protecting sensitive data at the column level, limiting what background functions are allowed to do, restricting access to live update channels, and removing photo uploads from the ambassador form to avoid anonymous upload risks.
Privacy-preserving analytics were built directly into the platform’s own database. Visitors are counted in a way that cannot identify them, using a daily changing code instead of any personal data. IP addresses are never stored, raw analytics data is automatically deleted after 90 days, and the codes used for counting are deleted after 180 days. Because no personal data is collected, the platform does not require a cookie consent banner.
Throughout development, I maintained clear product principles:
- Children’s safety as a non-negotiable constraint on every feature decision
- WCAG 2.2 AA compliance across all public and authenticated surfaces
- UK GDPR compliance with lawful bases documented per processing activity
- Opt-in for all data sharing, defaults are private
- No external analytics, telemetry, or third-party scripts without explicit approval
- Sensitive data protected
Result
STEM Demystified is now a live, publicly accessible platform at stemdemystified.com, featuring:
- 40+ role model profiles across 19 countries and six continents, with individual story pages
- A 3D interactive globe rendering real country shapes, two-category pin clustering, and hover/tap popups
- Public ambassador showcase listing approved STEM Ambassadors with country, discipline, tagline, and virtual talk availability
- Schools and Educators access to an ambassador directory, Ask a STEM Professional Q&A, and classroom resources
- Charities and Non-Profits directory with admin approval workflow and Technology Books for Children as featured charity
- Wall of Inspiration for children to submit drawings about STEM careers, with admin moderation before public display
- Full multilingual support in English, Spanish, Brazilian Portuguese, and German (including localised STEM acronyms)
- Progressive Web App support: installable to home screen on iOS and Android with no app store required
Wrap-Up
STEM Demystified began as a clearly defined gap, real stories that weren’t reaching the children who needed them, and became a fully realised platform through structured product thinking, deliberate privacy design, and continuous iteration.
It demonstrates how strong product ownership, domain expertise, and the effective use of AI tools like Claude and Lovable can deliver a complex, multi-user, multilingual platform that puts the safety and experience of children at its centre.
More than anything, it is proof that representation matters, and that the tools now exist to make it reach further.