My System
Voice Legacy is a production AI platform with real users and real revenue. I built it and run it without hiring anyone. People keep asking how, so I wrote it down.
Why Solo?
For the First Time in History, It's Possible
A year ago this wasn't feasible. Now I can build production software, run marketing, handle support, and manage finances—all with AI as my multiplier. The tooling finally caught up to the vision.
Iterate fast, then scale
I can move faster alone than any team weighed down by coordination overhead. I want to get traction first, prove the model works, and then hire or find a cofounder after raising funds. The order matters.
This one is personal
This isn't a side project. My father died and I couldn't capture his stories. That loss drives everything. I know how this product should work because I needed it and didn't have it. No hire is going to care as much as I do, not yet.
Development
How I work
I use VS Code when I'm editing files myself and Claude Code when I let things run autonomously. Either way, I stay in concept-space — describing features, reviewing code, and making architectural calls while Claude handles the syntax. I've made 647 commits in 7 months, and I haven't written a for-loop by hand since October.
The numbers
Six AI agents orchestrate the conversations, extract the memories, and write the stories. I built full auth, billing, a voice pipeline, and semantic search — 107 features across 12 categories, shipped solo. For most of my career I wrote requirements; now I write the code too.
QA & Evals
LLM eval framework
I built a custom evaluation system for the AI agents. It runs persona-based test scenarios with fake users like Maria Santos and James Mitchell, then scores them on a 100-point rubric across 9 dimensions. An LLM judge grades narrative quality, and a voice authenticity check verifies that James Mitchell actually sounds like he uses Southern Black vernacular.
A full eval run takes 15-30 minutes and costs about $1, or 45-60 minutes and $5 if I'm running Opus.
Automated Reviews
Two cron jobs run every morning while I sleep:
Picks next section of codebase, flags bugs, dead code, refactoring opportunities. No changes—just documents issues for me to review.
Runs security audit skill, checks for vulnerabilities, documents findings.
Design
Tools
- →An opinionated component system I built in code
- →Tailwind CSS for all styling
- →Mobile-first responsive design
- →No Figma — the design happens in code
Process
I built a strongly opinionated component system. Everything lives in code, not in design files nobody maintains. When I need a new component, I describe it to Claude, tweak the output, and ship.
Marketing
The Content Pipeline
I trained the voice model on 40+ samples of my own writing, including Substack posts, emails, and even Reddit comments. The output sounds like me instead of corporate AI slop.
Infrastructure
- →n8n self-hosted on pallas (Docker)
- →Twitter posting: working
- →LinkedIn/Facebook: debugging (it's always OAuth)
Total marketing infrastructure cost: $21/month plus the time to set it up.
Finance & Billing
Stripe Setup
- →Customer auto-creation on signup
- →Subscription tiers: Free / Pro
- →14-day trial support
- →Webhook handling for subscription events
Accounting
Wave (free) for bookkeeping. Mercury for banking.
Monthly Operating Costs
Support
How It Works
The Approach
The first layer is OpenClaw, which is an AI assistant that monitors in-app feedback, triages issues, and responds or escalates. The complex issues come to me.
Current volume is effectively zero since I'm pre-launch, but the system is ready.
I Can Teach You This
I consult on AI-native operations for founders who want to ship faster, PMs who want to become builders, and companies that need to figure this out before their competitors do.
