The Timeline: 30 Days, 1,293 Commits
Read This First
This is the origin story. Every other deep dive on this site covers a single project. This one covers all of them - when they were built, how they connect, and what the arc looks like when you zoom out.
The raw data comes from git histories across 26 repositories. The narrative comes from commit messages, project documentation, and the kind of clarity you only get in hindsight.
The Setup
In January 2026, I was a senior developer at a clinical diagnostics lab. Then I wasn’t.
The honest version: I helped a colleague analyze some data nobody asked me to analyze. It felt good. Not “productive” good. Good-good. Like remembering what the work was supposed to feel like. And then instead of opening LinkedIn, I opened a blank folder.
The question wasn’t “what should I build?” The question was “can I build anything real without a team, without a runway, without permission?” The answer turned out to be: a lot, actually.
Act 1: Building Season (Feb 8-24)
Week 1: The Framework Is the Product
February 8 - first commits. A friend had nudged me about solar education. I’d gotten a $50K quote to put solar on a cabin in Welches, a few months earlier. I had opinions. I started writing.
February 12 - the big bang. I needed a site for the solar content and a portfolio site for myself, and I didn’t want to build two CMSes. So I built one - an Astro-based framework called Loomwork - and immediately forked it into both sites. The framework and the sites were born in the same afternoon. Same initial commit across 8+ repositories.
February 13 - built the first version of Codenoscopy. An AI code review tool where you pick a reviewer persona and get feedback in their voice. Started as a joke. Still funny.
Week 2: Content as Code
February 15-18 - built two Home Assistant integrations: a peak demand shaver and a time-of-use metering system. One sprint day each. If you have solar panels and a Home Assistant setup, these actually save money.
February 21 - the content pipeline crystallized. The solar education material wasn’t just going to one site. It was structured to flow through prompts into different formats (web, PDF, presentation) with different tonal variants for different audiences. Same facts, different voice. One source of truth, many outputs.
This was the first time I used AI not just to build software but to transform content at scale. The pattern would show up again.
Week 3: Sprint Energy
February 22 - 35 commits on Codenoscopy. Built the entire interactive experience in one session. My commit message at 9 PM: “Can we go home now?”
February 23 - not a code day. A brain dump day. One session with Claude where I poured out 33 documents’ worth of accumulated notes - recruiter calls, project ideas, family logistics, business checklists, therapy appointments, a Reddit post that accidentally started a solar education community. The AI organized it all.
That session ended with a line I didn’t realize was important at the time: “This is the workflow. This is what Rejog needs to do.” More on that later.
February 24 - Loomwork grew up. Major mobile responsiveness push, framework-level improvements. It stopped being “the thing my sites run on” and started being something other people could use.
Act 2: The Crucible (Feb 25 - Mar 1)
This is the week that changed everything. Multiple projects born, one spectacular failure, and a methodology that emerged from the wreckage.
February 25 - Migration & Organization
Everything moved to GitHub. Repos organized into project families. A daily workflow system started taking shape - morning reviews, mid-day updates, structured intake of voice memos and notes.
February 26 - Sidefire in a Day
Built a complete group chat app in one day. Sidefire: WebSockets via Durable Objects, push notifications, emoji reactions, media uploads, phone number auth. 22 commits. It works. People use it.
The lesson: a single person with AI agents can build a real product - backend, frontend, real-time infrastructure - in hours, not weeks. But only if the scope is contained.
February 27 - Hubris
Fresh off the Sidefire high, I started the rewrite. Could an AI team build a full monorepo? React PWA, React Native mobile, admin dashboard, Cloudflare Workers API? Phases 1 through 5 done by 5:48 PM. Everything was going great.
February 28 - The 69-Commit Day
This was the most intense day across the entire timeline. Sixty-nine commits on one project. Most of them were fixes for problems created by earlier fixes.
March 1 - Thanks for the Memories
At 9:47 AM, I committed “chore: mothball sidefire2 - thanks for the memories.” Eighty-nine commits in thirty-six hours. Done. Not finished - done.
Then at 10:10 AM: “fix: do NOT delete sidefire-images R2 bucket - shared with v1.” Even in death, the rewrite was still dangerous.
The failure taught me the most important lesson of this entire timeline: an AI agent without a contract will build forever. The problem wasn’t the code quality - the agent wrote correct code. The problem was that nobody defined when to stop, what to verify, or how the pieces should fit together. The agent optimized for shipping features. It should have been optimized for shipping a product.
That afternoon, the daily workflow system had its biggest day: 45 commits. Voice memos processed. Lessons documented. The process of documenting what went wrong - honestly, with specifics - turned into the process of defining what should go right.
Act 3: The Synthesis (Mar 2-9)
March 2 - The Interview Protocol
I wrote the Interview Protocol: a structured conversation that produces a project contract in 20 minutes. Seven stages - from “what’s the one-sentence pitch” to “what does done look like.” The output is a WORKFLOW.md file with phases, gates, verification steps, and exit criteria.
To test it, I stripped Boopadoop (shared AI canvas - visitors describe what they want, Claude generates a live HTML site) to bare bones. Just a spec and a workflow contract. Found 10 gaps in the contract. Fixed them. Rebuilt from scratch using the new methodology. First project built under the process.
March 3 - The Actual Thing
Loomwork 2.0 features landed: themes, reader controls, the longform page template you’re reading this on right now.
And then the real work started. Using the interview protocol, I started the project everything else has been leading to. The methodology that was designed for web projects worked for iOS without modification. The contract was the same shape. The gates were the same gates.
March 4-6 - Steady State
Daily workflows running. The main project advancing under the process. Throughlinetech.net launched publicly on LinkedIn. Boopadoop fighting SMS delivery battles (34 commits in one day wrestling with Twilio 10DLC rate limiting).
March 7 - Pipeline Day
The iOS app got a flashlight feature from zero to TestFlight in 73 minutes (27 commits). The backend went from dormant to real - six pipeline phases implemented in a single afternoon session: ingest, LLM abstraction, cost tracking, triage, batching, context assembly, item creation, cascade evaluation. And at the end of the day, the emergency button feature was born.
March 8 - The Second 69-Commit Day
The backend became a real product: entity graph, confidence routing, onboarding interview API (three modes - structured, conversational, hybrid), 70 new tests, rate limiting, emergency features through the night. The iOS app got an emergency alarm system - dual-camera video recording, Bluetooth audio routing, custom alarm recordings, five flash rate modes.
The interview concept crossed over into the product itself: the onboarding flow IS an interview. The methodology and the thing it was meant to build started converging.
March 9 - Deploy and Lock Down
The backend deployed with 560 tests and 68 API endpoints. The iOS app got a round of fixes and an implementation plan for what’s next. And the portfolio site locked down its internal pages behind server-side auth - the last commit of the sprint: “lock things down.”
The Connection Map
Everything feeds the main thing. But the connections are specific:
Loomwork is the infrastructure layer. It powers every public site - this one, Johnny Solarseed, the solar variants. One framework, many faces.
The solar ecosystem tested the content-to-output-streams pattern. Same source material, different voices, different formats. That pattern applies far beyond solar education.
Sidefire v1 proved what one person and AI agents can build in a day. Sidefire v2 proved what happens when you don’t set boundaries on what gets built. Both lessons were necessary.
The Interview Protocol was born from the v2 failure. It now governs every new project. The contract is the artifact - portable to any agent, any runtime, any IDE.
Boopadoop was the first test of the new process. The main project was the second. The process worked for both.
The daily workflow system is the connective tissue. It captures context across all projects, processes voice memos and notes, and produces daily briefs that restore state instantly. Every morning starts with “where were we?” and gets an answer in under two minutes.
By the Numbers
| Metric | Value |
|---|---|
| Repositories | 26 |
| Total commits | 1,293 |
| Calendar days | 30 |
| Live websites | 6 |
| HA integrations | 2 |
| CMS framework sites | 8+ |
| Biggest single-day, single-repo | 69 commits (the one that taught the lesson) |
| Busiest combined day | 69 commits across 3 repos (Mar 8) |
| Backend tests | 560 |
| API endpoints | 68 |
| Commits at 3 AM or later | More than zero |
The Throughline
The portfolio answers “can this person actually build things?” so that when the main project ships, the answer is already obvious. Every deep dive on this site is a real project. Every one of them informed how the next one was built. The methodology got better with each iteration. The tools got sharper. The scope got tighter.
Thirty days. One person. AI agents. Portland, Oregon. February to March, 2026.
Not bad for a month.
This page was generated from git histories across 26 repositories. It can be regenerated anytime to reflect new activity.