ROOOD

Role: Lead Product Designer

Duration: 2 months

Mynu app screens: goal selection, home, and DNA results

The rise of artificial intelligence changed the rules of digital product design. Methodologies that for years shaped my work, like the double diamond, started to fall short next to tools that can generate in hours what used to take weeks. Integrating AI into the process stopped being a curiosity and became a necessity.

In that search I came across the Stingray model by Board of Innovation, which proposes a new way of understanding collaboration between people and AI. This case shows how I brought that model into a real project: turning research findings into a functional prototype, testing it with users, and using it to define an MVP.

Board of Innovation’s Stingray model: Train, Develop, and Iterate phases, from AI-led to human-led work, ending in concepts validated for desirability, viability, and feasibility

Client Context

Mynu is a Chilean personalized nutrition service that analyzes each person’s DNA from a saliva sample and delivers food and habit recommendations through an app.

At the time of the project, the service was already operating, but users perceived it as a genetic report, even though it offered much more. Its goal was to go one step further: to become a nutrition coach that accompanies people in their daily lives.

The Mynu sample collection kit: box, tube, swab, and instruction guide
A couple cooking in their kitchen while checking a phone

The Challenge

A valued but little-used product. The prior research, which combined benchmarking, expert analysis, analytics, surveys, and interviews, confirmed a retention problem with a clear cause: users valued their genetic results but couldn’t turn them into concrete habits. The app informed, but it didn’t accompany.

The numbers reflected it:

  • Only 0.13% of the people who viewed the product on the website completed the purchase.
  • The app had 11.7 daily active users against 73 monthly ones.
  • At least half of the new App Store downloads were uninstalled within the same period.

A process that made people wait. In a traditional flow, going from findings to a testable prototype takes weeks. During that time, neither the client nor the users can experience the proposed improvements, and wireframes only offer a limited approximation of the final experience.

Study results are confidential.

The Solution

To respond to both challenges, Agencia Match offered a new service: turning research findings into a functional prototype generated with AI, usable and interactive from the earliest stages. As Lead Product Designer, it fell to me to give shape to that promise and turn it into a concrete working process.

The challenge was defining how to integrate AI without losing design judgment: which tasks to delegate, which decisions to keep within the team, and how to make sure what was generated responded to real problems. To structure it, I took the Stingray model as a base and adapted it to the project’s reality in three phases:

A driver at the wheel with a phone mounted on the dashboard

01

Train, Research as training.

I turned the findings, user personas, and user journeys into the context to train an agent on the project’s goals.

02

Develop, Generative exploration.

I used AI to generate interfaces and alternatives in parallel, widening the creative space before converging.

03

Iterate, Validation and prioritization.

I evolved the prototype through six main rounds of iteration, we tested it with users, and we used it to define a viable MVP.

The prototype addressed all the findings of the report and went deeper into goal setting, meal logging, the organization of the DNA results, the findability and comprehension of tips by category, and the connection with wearables.

Process and My Contribution

The process in three phases. Train: analysis of the research results, goal definition, training the agent in ChatGPT. Develop: AI tool benchmark, initial generation of the app and the website, exploration of alternatives. Iterate: six rounds of iteration, user testing, initiative prioritization and MVP definition. I led all three phases

Phase by Phase

lead

Train, From findings to context

The Stingray model holds that, before designing, AI must know the problem as well as the team does. That is why my starting point was to train an agent in ChatGPT with the project’s full context: the retention problem, the prioritized findings, the user personas and their journeys, and Mynu’s business goals. This step kept the generated interfaces from being generic solutions and ensured that every proposal responded to a real problem detected in the research.

Uploading the two research reports to the agent: the Mynu user research report and the prioritization workshop report
The agent’s prompt structure: agent role, context, objective, design principles, iterative working method, rules for generative AI UI prompts, and response format

Develop, Explore before converging

In this phase, the model proposes opening up the range of solutions as wide as possible, leaning on AI’s ability to generate many alternatives in a short time. Before committing to a tool, I compared Lovable, Figma Make, and Uizard based on the quality of the screens they generated. Lovable performed best and was the one chosen.

My first instruction was deliberately broad: build an app that solved the problems identified. The goal wasn’t to reach a final answer but to quickly get a functional base to iterate on and to explore paths that would hardly have emerged with wireframes.

Four app screens generated with Lovable: kit activation, home, meal logging, and progress
Four app screens generated with Figma Make: welcome, home, daily log, and personalized plan

Iterate, From the general to the particular

With a first functional version in place, the work became progressively more specific: first at the level of flows and then of screens, across six documented rounds that included targeted iterations on key sections.

We tested the prototype with 10 users in semi-moderated sessions, and all of them completed the assigned tasks. Because it was a realistic, interactive tool, participants had an experience much closer to a real product, which let us gather richer and more complete insights than with a traditional prototype.

Finally, the prototype played a role that went beyond validation. Since it wasn’t feasible to implement all the improvements at once, it became a tool for sifting and prioritizing initiatives. Being able to see and try each solution made the conversation with the client easier and let us converge on an MVP that balanced what users expected with Mynu’s technical and commercial limitations. This is where the Stingray model shows its greatest value: validating desirability, feasibility, and viability at the same time, rather than one after another.

Onboarding, turning the wait for results into the start of the relationship with the user.

  1. Goals tailored to the user. Users choose up to three goals (health, energy, weight, performance, eating more simply, or caring for their family), aligned with the user personas’ motivations.
  2. More context, better content. Data such as age and activity level make it possible to deliver relevant recipes, habits, and recommendations from day one.
  3. Education from the start. Screens that explain why DNA matters in nutrition reinforce the value proposition before the results arrive.
  4. Sample traceability. The flow shows each stage of the process and its estimated time, reducing the uncertainty of the wait.
  5. Early access. Users can explore the app and log meals before receiving their results, building adherence from day one.
The final onboarding: welcome, goal selection, personal data, DNA explainer, kit activation, and what happens while waiting for results

Home, turning the nutrition plan into daily actions and conveying support.

  1. Visible progress. The Mynu Score and nutrient tracking replaced the initial goal list with a more motivating, easier-to-read indicator.
  2. Suggestions with a why. Personalized recommendations explain how they relate to the user’s genetics and translate into concrete actions.
  3. Quick logging. Shortcuts to log meals, water, supplements, and exercise, the most repetitive actions of daily use.
  4. Meals with context. Each logged meal shows its nutritional contribution.
  5. Conditional wearables. Activity and health data appear only if the user has connected devices.
The home screen across iterations, from the first version generated with Lovable to the final result with the Mynu Score

Tu ADN (Your DNA), moving from a dense report to a clear reading of what the results mean and what to do with them.

  1. Progressive disclosure. Information is organized in three levels: profile summary, key findings, and the detail of each one.
  2. Categorization that makes sense. Results are ordered by:
    1. Priority: optimal, moderate, or needs attention.
    2. Nature: nutrition, metabolism, sensitivities, and wellness.
    3. Application: eating and wellness guidelines on what to prioritize, moderate, or avoid.
  3. Education in context. Each finding explains what it means, what to do, and which foods are recommended.
  4. From data to action. A weekly plan and a shopping list turn the results into concrete decisions.
The Tu ADN section across iterations: first version, intermediate iterations, and final result with the profile summary, key findings, and categories

Try the prototype

The functional prototype generated with Lovable that we tested with users and used to define the MVP. It runs live: tap through the onboarding, log a meal, and explore the DNA results. The interface is in Spanish, as in the real product.

Open in a new tab

Learnings and Conclusions

  • Context quality defines result quality. A well-trained agent, fed with the research, produces solutions connected to users’ real problems.
  • From the general to the particular. Starting with broad instructions and then going down to flows and screens proved more efficient than asking for detailed screens from the start.
  • Choose the tool with evidence. The AI ecosystem changes fast; comparing options avoided adopting the first one available.
  • A realistic prototype improves testing and decision-making. The more it resembles a real product, the more honest users’ responses are and the easier it is to prioritize with the client.
  • A window into the product’s potential. The prototype lets the company see what it can become. From that vision, and with realistic planning that crosses users’ needs with the business’s possibilities, the MVP, the backlog, and the releases are defined more clearly, concretely, and soundly.
  • The designer’s role expands. With AI generating alternatives, the value of design lies in giving direction, curating options, and deciding what to build.

This project was a successful process and showed me that AI doesn’t replace design judgment: it speeds it up, expands it, and makes it easier to share.

Results

-35%

Time to build a prototype for testing, compared with similar processes on projects of Mynu’s scale.

10%

Efficiency gain in the ideation process.

100%

Of participants completed the test tasks (10 users, semi-moderated sessions).

Prioritized MVP

aligned with users’ expectations and the business’s technical and commercial constraints.