AI Bootcamp Day 3: Building, Testing, and Thinking Beyond the Bootcamp
November 08, 2025
The final day of South Florida Tech Hub’s AI Bootcamp wrapped up with a mix of reflection, experimentation, and project demos that proved how much ground we covered in just three sessions.
Lessons from the Roundtable
We started the morning with a round robin of successes and roadblocks. A recurring theme: running out of free credits on the AI tools we were using. Many of us hit strange errors mid-project—sometimes on every response. It became clear how important it is to understand the kind of code or content you’re generating.
If you don’t know what “good” looks like, you might not realize when an AI gives you bad output. Garbage in, garbage out—and sometimes, garbage looks convincing.
Another strong takeaway: there’s no perfect tool. What works for one project or person might fail for another. The best approach is to describe your project to ChatGPT, Claude, or Gemini, ask questions about which tools might fit, and then experiment.
As Damian’s dad put it, “Perfection is the enemy of success.” Get something working first. Refine later.
A Quick Reality Check: Data Privacy
We also talked about privacy and data retention.
- All LLMs save your prompts indefinitely.
- Free plans have no confidentiality—no contract, no protection.
- Pro plans include contractual obligations, but your prompts are still stored.
It was a sobering reminder that even when AI feels like a personal workspace, it’s still a public system.
Lunch (and Pizza) Before the Show
After lunch, we kicked off project presentations—the culmination of everything we’d been experimenting with since the first Saturday.
Highlights from the Projects
The creativity on display was astounding:
-
Mark Tennenhouse built a Piano Learning System Origin—a game that teaches piano by using it as a controller. Think Flappy Bird, but you keep the bird aloft by playing the right notes. He re-engineered the controls so music drives motion.
👉 marktennenhouse.github.io/html-games/flappy-bird/index.html -
Patty Tennenhouse (Mark’s wife) used ChatGPT and Cursor to create a KDP Cover Generator for Amazon print-on-demand. She only asked for a few covers—but Cursor built a full-blown cover generator app. She accidentally worked in agent mode instead of ask mode, which made the AI more autonomous. A happy accident.
-
Ron Vélez presented The Home Insta Lab System (THILS) — an AI-powered at-home diagnostic tool that collects and analyzes health data, then shares it securely with your doctor. He even built a HIPAA compliance “agent” inside Cursor to review his code. The marketing video? Generated with HeyGen and scripted by Cursor itself.
-
Juan built LinkBridge, an email summarization tool that condenses long email threads into key points, action items, and unanswered questions. It runs on ChatGPT 4 Mini through an OpenAI API key—essentially, a smart inbox assistant.
-
A medical intake agent project streamlined patient check-ins for diverse, multilingual clinics. The system collects patient data conversationally and suggests CPT codes based on visit type—built with Cursor and GitHub Spark.
-
Jade, a grad student at the University of Miami, built LegalMind Fatigue Monitor — an AI-assisted decision support app that tracks mental fatigue and can sync to wearable devices. She used Copilot to generate and test the app’s structure.
-
Luis, a computer science major at FIU, created Confusion to Clarity: Career Resilience — an AI-driven career development app that helps users analyze skills, identify gaps, and understand how automation might affect their chosen field. He built the site in ten minutes on Replit and used Sora AI from ChatGPT to enhance images for his project. He originally took the job expecting IT work, but it turned into a graphic design role — and thanks to Sora, his AI-generated images impressed his boss. Luckily.
-
Another participant, a contractor for Mr. Beast, showed how far low-code AI tools can go. Using Claude, he built a fully responsive website with parallax effects and a persistent logo header, hosted it on Firebase, and even experimented with a voting feature for Beast Games 2. The most impressive part? He completed the whole project in about a week without traditional programming. Claude guided him step by step — including deployment instructions.
-
Lokesh built InnerLift, a “gym mindset buddy” that pairs users based on training style and emotional compatibility. Using Copilot, he created a UI that generates emotion pie charts and match scores from user input.
-
Brandon Ray’s Introflo.io tackled a real-world gap in behavioral health referrals. His AI helps solo mental health practitioners form “warm handoff” connections with other providers based on specialty, insurance, and proximity—something existing systems like Salesforce or Trillium can’t handle gracefully.
-
Michael Medina shared Cyclopedia, his passion project turned professional-grade hurricane database. With help from Cursor, he migrated it to Next.js for server-side rendering and performance optimization.
👉 github.com/mikematics22800/Cyclopedia-Next.js -
Another attendee closed out the presentations with a reflection that summed up the spirit of the weekend: “All those bad ideas I shelved for years—AI turned them into weekend projects.”
The Big Picture
That final thought really captured the theme of the bootcamp. It wasn’t about becoming AI experts—it was about building momentum. Trying things, breaking things, learning what works, and realizing that progress doesn’t require perfection.
AI is no longer something we read about. It’s a toolset we can wield, experiment with, and shape to fit our ideas. And that’s the real shift.
Bootcamp Complete.
Three days, dozens of tools, and countless “aha” moments later—what started as curiosity turned into capability. The best part? Every single project proved that creativity, not perfection, is the real entry point into AI.