How I Used AI with a Team at the NASA Hackathon to Build an MVP in 32 Hours
I recently took part in the biggest hackathon in the world in the world, the NASA Space Apps Challenge, an intense event of 32 hours to build a solution with real impact, with no commercial purpose, aimed at people and the environment.
How the process unfolded
No one on the team knew each other. So we started by talking to build empathy and understand each person's skills: how each of us could contribute and how that connected to NASA's challenges.
We identified skills in software development, business, hardware, space science, design, and product.
We held an initial brainstorm to choose a real problem to solve, all in that very first conversation.
We defined which challenge of the event our idea fit. From there, we discussed what the expected user experience would be, and what the person could do when entering the system.
We created a simple sketch, with just four steps, so anyone could get the information they needed quickly.
Meanwhile, part of the team used AIs (Claude and Genspark) to process all the ideas from the conversations, the brainstorm, and the challenge description. Everything happened in parallel: while we were drawing the journey on paper, another teammate was generating visual ideas in Genspark.
Later, a teammate wrote the pitch (with help from ChatGPT). Everything was shared in a WhatsApp group. Then another teammate took that pitch and generated a detailed documentation in Genspark about what the system could do.
I started to refine the visual ideas in FigJam, creating prompts for each screen. Another teammate, using Claude, structured a complete prompt with the four screens, describing what each one should solve.
Meanwhile, the developer on the team was studying with Claude how to integrate images and prompts so that another AI could interpret them and generate the front end.
With everything structured, I prepared a document with the prompt for the four screens and the corresponding images (Screens 1, 2, 3, and 4) to be used in v0.dev and in Lovable.
In the first version, v0 did better, so we chose to continue with it, hosting the project on Vercel. Development would be done in Cursor, integrating the front end generated by v0 with the back end and the NASA APIs.
The back end was challenging, because we were still learning how to use Cursor.
At night, we realized that we had not defined a method for updating the front end in v0 without breaking the code in Cursor. When we updated the front end, the code in Cursor broke, and we lost time rolling back versions.
We learned that we should have defined this flow at the very beginning, understanding how the AIs would complement each other. That way, we could work on front end and back end separately before joining everything.
The integrations were built in Cursor, with constant help from Claude (which, of course, froze and hallucinated several times). Meanwhile, the rest of the team was polishing the presentation pitch.
Using Claude and FigJam, I created some personas to help focus on who we were trying to solve problems for. This work shaped the narrative we would tell about the project. I made the images using ChatGPT 5.
To avoid new errors and make use of credits I had, I decided to rebuild the entire flow and the interfaces in Lovable. But first, I wanted to improve the user experience as a whole.
As a designer, I reviewed all of the v0 screens in FigJam, created a screenflow, and did a critical assessment of what needed to be improved.
In FigJam, I cut steps and made communication simpler and more practical, so the user could decide clearly how to leave home for outdoor activities without compromising their health.
In the final moments, we focused on filling out the project form on the NASA website.
At the end of the second day, we presented the project to the event's consultants and received many valuable pieces of feedback.
We hurried to fix the points raised and, finally, delivered the complete project.
The final interface, built entirely in Lovable: https://responsive-grid-recreation.lovable.app/
We also created a landing page for the project, also in Lovable: https://responsive-grid-recreation.lovable.app/
What we did well 💪
Speed in decision-making
Agile collaboration between design, engineering, and product
A simple, clear, and collaborative user journey
A focus on solving a lean and functional flow
Using AIs to generate, structure, and validate ideas
Building a front end in minutes with AI tools
Main lessons learned 🌱
Spend more time assessing the problem and understanding who it is for (persona, journey, and moment).
Define one person responsible for coordination on the team, and make sure there is focus and cohesion.
Study the challenges in more depth before starting to build.
Look for problems that are real, clear, and measurable.
Have clarity about who is affected and how the solution creates impact.
Invest more time framing the problem and showing its social or environmental impact.
Build a more powerful narrative about how the solution improves people's lives and the environment.
Understand that AIs hallucinate when they receive too much data.
Learn to write better, more contextual prompts.
Structure the workflow between AIs, front end, and back end from the start.
Comments