The numbers tell the story better than I can: 2,652 commits across 76 new projects, spanning 14 programming languages, with 11 reaching production maturity. This was my year of vibe coding at home.

Q1: The Business Validation Phase

The year started with a familiar pattern: chasing business ideas.

blueskyreels (136 commits, January) was an exploration into Bluesky and video players—could there be a business here? The audience felt off. Too risky. Abandoned completely.

homerunleague and CommunityRefugeeAid came from other people’s problems and my curiosity with PRD-driven development. Spec it out, build it fast, see if it works.

But the real story of Q1 was finally shipping BicycleRadar (bikesafetyradar3, 78 commits). I’d been pivoting on this cycling safety project for ten years. Ten years of “the architecture needs to be just right.” Ten years of “the hardware integration needs more research.”

I shipped it. Free. And realized: I still don’t really need to make a business out of this.

April: The Inflection Point

April was my quietest month—only 55 commits. But it was the most important.

I was at a crossroads: start a thorough effort to make an income from side projects, or just build for vibe coding’s sake? The answer came slowly, then all at once.

Just build.

The Gemma 3n Catalyst

Then June happened. 864 commits in a single month.

Google’s Gemma 3n with multi-modal input was a complete game-changer in how I think about AI and useful applications. The Kaggle hackathon deadline pushed me to ship, but the technology opened doors I didn’t know existed.

Quarter	Commits	Repositories	New Projects
Q1	273	8	3
Q2	923	16	15
Q3	618	47	46
Q4	838	12	12

Q2 alone produced nearly 35% of my yearly commits. That’s not a gradual ramp-up. That’s a capability unlock.

June: The Explosion

Three major projects emerged from the hackathon frenzy:

HeatSentry (351 commits) — Humidity-related readings and environmental monitoring. A complete Dart application built in weeks.

Dronevade (292 commits) — Drone detection with multiple layered analysis: type, category, payload identification. Technically neat, but couldn’t pivot to a business. The detection patterns were fascinating though—cars and drones share more in common than you’d think.

LoopSweep (114 commits) — A brain-cleansing fun idea with AI-generated UI. Pure experimentation. Abandoned, but the Swift skills carried forward.

And then the one that mattered most:

LiveCaptionsXR (170 commits) — I realized I could try spatial intelligence with closed captions for my own hearing impairment. This wasn’t a business pitch—it was solving my own problem. The project didn’t win the hackathon (it doesn’t quite work exactly how I wanted), but it’s still a useful idea in need of refinement. This one continues.

FarmhandAI was a secondary hackathon submission I had lined up—agricultural AI seemed like natural territory for edge computing.

The Experiments Continue

August brought aipartpicker (87 commits)—a way to help me organize how I’d build a local AI system. Abandoned, but the research informed later work.

sealfold-dashboard (25 commits) was an attempt to challenge business suite applications with provenance tracking. Multi-agent collaboration dashboards felt like the future.

September saw BananaBread (103 commits)—an API wrapper to experiment directly with AI APIs. The nano banana model was fascinating to work with. Abandoned, but could be a business with polish.

NavierStokesProof (23 commits) was an attempt to recreate a mathematical paper. Not even close. 😅 Abandoned. Some problems need more than vibe coding.

October brought AntifascistFunBrigrade (115 commits)—inspired by a book on challenging authoritarian governments with fun, nonviolent protests. The backend had AI image generation panels with pre-prompt selectors. Creative tooling for creative resistance.

The XR Pivot

By late 2025, I realized spatial computing with XR might be the next frontier. Google’s Android XR demos in May were compelling—but how could I use it?

The answer: build upon LiveCaptionsXR. Evolve it into a system for controlling robots at home. A trainer and PC replacement for spatial computing.

ContinuonXR (612 commits) consumed November and December—536 commits in December alone. It’s an XR system that grew directly from the accessibility work, expanding into broader spatial computing applications.

ContinuonAI represents something more ambitious: proving that general AI for useful embodied robotics is feasible with under $600 in parts and a shared “brain” model that can work across robots while retaining local memories. No transformer-based LLMs required.

SlopMusic (42 commits) was a one-day experiment with music generation. When you can build that fast, why not try everything?

SchweitzerPTA was a neat workflow experiment: Google Stitch → Jules → Antigravity → Gemini CLI with conductor. The toolchain for vibe coding keeps evolving.

Why Home, Not Work

I keep my professional vibe coding minimal. Experiments are best left for home use.

At home, I choose the architecture, languages, frameworks, and tooling without compromise. I chase rabbit holes. Dart one week, Rust the next, TypeScript after that.

• Python: 11 repositories
• TypeScript: 11 repositories
• Dart: 11 repositories
• JavaScript: 11 repositories
• PowerShell: 8 repositories
• Plus Swift, HTML, Astro, Go, C, C++, Rust, and MDX

Professional codebases need stability. Home labs need freedom.

Do The Tradeoffs Matter?

Do the costs and tradeoffs of AI tooling really matter for home projects?

They don’t—until they do.

Most of these tools work with free tiers or very reasonable subscriptions. I’m not running a business on most of these. I’m learning, experimenting, and occasionally shipping something useful.

The real tradeoff is mental: accepting rough first versions, accepting that projects will be archived after burst activity, accepting that the value was in the building.

The Cleanup Ahead

There’s a flip side to explosive productivity: repositories are growing. Some reflect unfinished and bloated states of rapid experimentation. Not every burst project needs to live forever.

It’s a good time to archive some of these. Some I may revisit (BananaBread could be polished). Some I may publish. Many served their purpose as learning vehicles.

The beauty of home coding is that there’s no pressure either way.

Meta: This Post Is Vibe-Coded

This very blog post demonstrates what I’m describing.

The analytics, the heatmap visualization, the project classifications—all generated by a Python script I built with AI assistance. It pulls from the GitHub API, classifies projects, generates narratives, and produces the SVG timeline above.

Want to run your own year-in-review? Check out the README:

• Complete repository analytics as JSON
• Visual activity timelines as SVG
• Blog draft templates
• Monthly recaps with narratives
• Individual project spotlight files

The script took an evening to build. It would have taken a week before.

Looking Forward

Seventy-six new projects. Eleven reaching production. A ten-year project finally shipped. An accessibility tool evolving into embodied AI for home robotics.

This isn’t about AI replacing developers. It’s about removing the friction that kept projects in limbo. It’s about matching the pace of building to the speed of ideas.

The trajectory is clear: from chasing business validation to just building. From accessibility tools to spatial computing. From single projects to an ecosystem.

2025 was the year vibe coding went from curiosity to workflow. ContinuonXR and ContinuonAI point toward what 2026 might bring: home robotics, embodied AI, and spatial computing—all built at the speed of vibe.

---

Generated with data from the GitHub Timeline Analytics tool—itself a product of vibe coding.