Side Projects: Making life easier with software
Come back to this - Jan. 2026 - Claude Prompt
You are an experienced app developer tasked with creating a detailed plan for building an app using an AI enabled IDE. Your goal is to provide a comprehensive outline that will guide the development process. Follow these instructions carefully to create your app plan. First, you will be given an app description and the target platform. Read them carefully:
Now, follow these steps to create your app plan: 1. Analyze the app description: - Identify the main purpose of the app - List the key functionalities required 2. Identify key features and components: - Break down the app into its core components - List any additional features that would enhance the user experience 3. Outline the app structure: - Determine the main screens or views needed - Plan the navigation flow between screens 4. Plan the user interface: - Describe the layout for each main screen - List any specific UI components or widgets required 5. Consider backend requirements: - Determine if a backend server is needed - Outline any database requirements - Plan for data storage and retrieval methods 6. Determine necessary APIs or libraries: - List any external APIs that will be used - Identify third-party libraries or frameworks that could be helpful 7. Outline testing strategy: - Plan for unit testing of key components - Consider integration testing requirements - Outline user acceptance testing criteria After completing these steps, provide your app plan in the following format:
Remember to tailor your plan to the specific requirements of the {{TARGET_PLATFORM}}. Consider any platform-specific design guidelines, performance optimizations, or features that should be incorporated into the app. Provide your detailed app plan based on the given app description and target platform, ensuring that it covers all the aspects mentioned above.
Copy the output (PRD) from Claude - to Stitch to get the HTML - Then you can export it to Google AI studio - then connect it to supabase (For backend)- then upload to lovable
Loop - Dec. 2025 - Sixth post - Link Here to Try it Out!
Big day for Loop. Had some downtime after the holidays so got to play around with it for a little bit inside lovable.
Summary here:
- I reworked the UI to feel way more interactive — lots of pop-ups, feedback, and moments that actually react to what you do. It finally feels alive instead of static.
- The biggest addition: Challenges.
- You can now challenge anyone in your cohort, wager up to 60% of your weekly points, and they’ll get a global alert wherever they are in the app. No real-time coordination needed — they come back later, get 30 seconds to answer as many questions as possible, and the winner takes it.
- Once both players finish, Loop delivers a result screen with some encouraging (and slightly rude) copy depending on whether you won or lost. And if you lose? You can instantly rematch, which keeps people in the app and fuels that “one more round” motivation.
I also made a conscious product decision here: Loop is staying cohort-focused. Turning this into a generic Trivia Crack clone would dilute what makes it different. Competition hits harder when it’s against classmates, not strangers.
On the backend, I added an admin layer:
- Students are auto-assigned to universities based on email domain
- Admins can create cohorts/majors under universities
- Users can be manually reassigned as needed
It’s starting to feel less like a prototype and more like a real system.
I also added a cohort chat and reorganized the dashboard so the “My Cohort” section gets more screen real estate. You can now see who you’re competing with at a glance and chat directly with people in your cohort. It definitely needs more refinement, but the social layer is starting to take shape.
Next up is tightening the community feedback loop — more real-time, visible signals like “Eli moved up to 2nd place” or “You dropped to 3rd”. Those small nudges are what make the leaderboard feel alive instead of just a static list.
More soon - I should probably include an explanation page so when users sign in they know what they're getting themselves into... like a quick video to show or a gif
Loop - Dec. 2025 - Fifth post
Tried out three new layout styles for Loop this weekend: the classic Duolingo look, a chaotic-but-fun Keith Haring vibe, and a super clean black-and-white cartoon mode.
It’s very obvious which one is actually usable… but honestly it was cool seeing how fast I could flip the whole app into different aesthetics. One prompt and the vibe changes completely.
Loop - Nov. 2025 - Fourth post
If you don't know what loop is - read from the first post below por favor.
I spent a few hours on Loop this weekend and here's some AI assisted Updates:
There are now actual choice paths when you sign up, multiple subjects, and real test questions. I even put some of the UX research practices into play (discussed below) added menus and buttons using “heuristic principles,” essentially, I cleaned up the chaos. Imposter syndrome.
The new idea is sort of a community mode.
When someone signs up, they can pick their school and cohort. Why? Because attention spans are cooked, and the only way anyone learns now is through games and competition.
So imagine telling your class: “Hey guys, sign up, join our cohort, and play me. Loser buys beers.”
Where it’s heading (Or at least I think)
Think Trivia Crack × Kahoot × Quizlet × bar trivia.
Challenge people. Go 1v1. Maybe even chat… I don’t know. Let me dream.
Still on the to-do list:
Automate content uploads with Zapier/AI wrappers
Add admin roles (should be easy… famous last words)
La Huerta Group Project - Nov. 2025 - Digital Experience and CRO
We built La Huerta, a hypothetical “farmers meet consumers” marketplace, and it was one of my favorite mini-builds this term. Two phases. First we mapped the whole thing out, personas, journeys, information architecture, and then turned it into a mobile-first Figma prototype (shoutout to Fynn my German buddy for carrying the design & idea, excited to see what he does with it) .
Then the second part is the CRO side: card sorting, heuristic reviews, user testing… basically finding every point where a user might get confused and fixing it, "you need to be humble with UX research" says my professor. We added a guest checkout for a quick win. Think about it next time you're booking something, why is that button placed there? Or why that back button.
We created a quick mobile landing page, which you can check out below as well
https://elijahlarson.wixsite.com/lahuerta
First Delivery Slide Deck & Figma Link - La Huerta Deliv. 1 Draft - Final.pdf
Second Delivery Slide Deck (Try out the QR Codes!) - La Huerta Deliv. 2 Draft - Final (1).pdf
More of this kind of work coming soon.
Loop - Nov. 2025 - Third Post
Quick update on Loop, my Duolingo-for-learning-anything experiment.
Last time I used ChatGPT + Lovable to spin up the first version (seriously, a few prompts and some “fix this” clicks and it was alive). Since then I’ve been thinking about how wild learning has become; AI quizzes, AI podcasts, AI everything. No excuses for forgetting SQL now.
I was building an AI agent to auto-grab new slide decks and feed them into Loop… but my N8N free trial ended and the whole thing instantly stopped making sense. So we’re rebooting that part.
But! Loop is up and running:
https://deep-learn-loop.lovable.app
More to come.
Duolingo Idea - Loop - Oct. 2025 - Second Post
Okay so I started with ChatGPT obviously to generate a prompt for lovable, a no code application builder that you can use with NLP. Take a look at the output below in chatgpt. Copy and paste the output into lovable, follow a couple of buttons to troubleshoot and fix bugs and voila, a no code app live with a similar interface to DuoLigno, but for SQL (for now). See screenshots below for the first iteration.
Now next I'd like to use a no code AI agent to retrieve new downloads (like a slide deck) and upload that into lovable, so then I can have relevant and timely information in the platform. It's called Loop by the way.
Using n8n to comb through LinkedIn job recommendations - Oct. 2025
Just an idea! LinkedIn Recs are great but what if I could use an LLM to actually pick them based off my resume. So practical right now in a master's degree. Look at the photo below, it's helpful sure but not specific enough. Am I qualified? Are they cool companies? We'll see once I build this agent.
Duolingo, but for more than just languages - Oct. 2025 - First Post
I love Duolingo's app. It's interactive, captivating and rewarding. All in all, it promotes learning in an engaging and perfectly annoying manner (from their notifications).
I also love when knowledge sticks, but I know it doesn't if I don't practice every day. For example, SQL and R from a data analytics course this summer. I loved learning it, but didn't practice it.
The idea? What if I could have an app loaded with slide decks, or case studies, or information generated from LLMs and have it feed into this skeleton that'll do the same type of learning as Duolingo.
Maybe it won't work, maybe I can use Lovable or Base44 for NLP app building. More to come here.
Zapier AI Agent - Sep. 2025
I built a Zapier project that takes all my newsletters from the day and turns them into one clean digest at midnight. The flow: Zapier searches my inbox for yesterday’s emails, grabs the links, runs them through Gemini for a quick summary, and then creates a Gmail draft titled “Here’s the news from yesterday.”
The main challenge was that early versions made multiple drafts instead of one. I fixed it by switching to a scheduled search at midnight and adding a storage step so the Zap only runs once per day.
Now I get a single, scannable summary instead of 10+ scattered emails. It saves time, reduces noise, and makes newsletters actually useful. It's also a little redundant, yes I know but it's cool and I'd like to explore more there.
Posture Sensing MVP with Teachable Machine - Jun. 2025
For this project, I wanted to test if I could build a quick prototype of a posture tracking tool. I used Google’s Teachable Machine, a no-code platform that lets you train machine learning models directly in the browser with just a webcam and a few labeled examples.
I recorded different positions—upright, slouched, arms resting differently and trained the model to recognize patterns like the angle of my shoulders relative to the camera and my arm placement. Once trained, the model could give me real-time feedback on whether I was sitting with good posture or not, as long as my camera is on.
The MVP is simple, but it works. It shows how easy it is to use accessible tools like Teachable Machine to build machine learning prototypes in literally 60 seconds, not weeks. From here, I could imagine extending the project into something more useful: reminders when posture slips, analytics on sitting habits, or even integrations with productivity apps.