DEV Community

Domonique Luchin
Domonique Luchin

Posted on

How a Structural Engineer Built a 6-Business AI System From His Phone

My day job is structural engineering. I check steel connections and review load calculations for oil and gas infrastructure across the Gulf Coast.

I also run six businesses. On my phone. Using AI agents I built myself.

Here is how that happened.

The Six Businesses

Load Bearing Capital — Real estate wholesaling. AI scrapes 14 Texas counties for distressed properties, calls homeowners, and qualifies leads.

Load Bearing Demo — Demolition contracting. AI monitors Gmail for RFQs, parses drawings, generates bids using GPT-4o, and sends them after Telegram approval.

Load Bearing Detailing — Structural steel shop drawings. AISC-compliant. The same engineering skills from my day job, monetized as a service.

Quiet Hours Valet — Valet trash for apartment complexes. AI receptionist handles inbound calls and books property manager demos.

Luchin Credit Repair — FCRA dispute engine. Automated 30-day dispute timelines and follow-up sequences.

Petroleum Noir — Mineral rights and landman services. AI scrapes Texas Railroad Commission for new drilling permits and reaches out to operators.

All six run on one $28/month VPS and one $25/month Supabase instance.

Why a Structural Engineer

Engineering teaches you to think in systems. Every structure has load paths, redundancy, and failure modes. Building an AI agent system is the same problem with different terminology.

Your data flows are load paths. Your retry logic is redundancy. Your approval gates are failure mode analysis.

The code itself is not harder than structural calculations. It is different notation for similar thinking.

Building From a Phone

I do not have a home office setup. I work from my phone most of the time.

Claude on mobile handles architecture decisions. Claude Code on the server executes the builds. Supabase MCP connects Claude directly to the database without me writing queries.

The workflow:

  1. I send a task to Claude: "Build a lead scoring function for Harris County properties"
  2. Claude writes the SQL migration and inserts it into claude_code_queue
  3. Claude Code on the server picks up the task and executes it
  4. Result comes back to Supabase. I query it from my phone.

I am the decision maker. The agents are the execution layer.

The Part That Surprised Me

Building the stack was not the hard part.

The hard part was building in public while working a full-time job and relocating to a new city for a lead structural engineering position.

The system has to run without me. That forced better architecture than anything else. If I had to touch it every day, it was not good enough.

Every agent writes its results to Supabase. Every failure gets logged. Every approval request comes to my Telegram. I check it once in the morning like reading emails.

The goal is not to work less. The goal is to do $10/hour tasks with agents so I can spend my time on $10,000/hour decisions.

What Is Next

RunPod GPU pipeline for AI avatar videos. YouTube and TikTok at scale, all six brands, fully automated.

The voice samples are in Supabase Storage. The face photos are there too. Coqui XTTS handles voice cloning. MuseTalk handles lip sync. FFmpeg handles the final render.

One command will produce a video for any brand on any topic in under 10 minutes.

Building it now.

Top comments (0)