I wasn't always just a sales rep.

At 14, I was teaching myself HTML from internet forums — building pages just to see if I could. At 17, I was deep in IRC communities, fascinated by how systems worked under the hood. Pure curiosity. I wanted to understand the mechanics of everything digital around me.

For years I automated everything I could in Excel with macros and VBA. If there was a repetitive task, I built something to eliminate it. That instinct never left.

But life took me in a different direction. I became a commercial representative in the packaging industry — selling corrugated cardboard, managing sales portfolios, negotiating contracts across three Brazilian states. Good career. Real expertise.

Along the way, I also participated in the full implementation of Winthor ERP (PC Sistemas) — covering accounting, inventory and sales modules. I didn't just use systems. I watched them being adopted, saw where users struggled, understood what made software succeed or fail in the real world.

The tech curiosity stayed dormant.

Until AI woke it back up.

The Problem I Was Living Every Day

My entire operation ran on spreadsheets.

Orders, quotes, invoices, commissions, client health, product catalog — everything scattered across dozens of Excel files. Every month I'd spend hours manually calculating commissions, cross-referencing data, building reports for management.

I knew exactly what the software needed to do. Fifteen years of living the problem gave me that clarity.

I just needed the tools to finally build it.

How I Actually Build Software

I want to be honest about this: I use Claude and Cursor as my primary development tools.

I'm not the developer who memorizes syntax and builds from scratch in a blank file. I'm the developer who understands the problem deeply, architects the solution, and uses AI to accelerate the implementation.

That teenage instinct for understanding systems? It turns out that's exactly what AI-assisted development requires. You can't just prompt your way to a production system. You need to think in architecture. You need to understand why things break. You need to know when the AI is confidently wrong.

Fifteen years of business domain knowledge + ERP implementation experience + decades-old curiosity about how systems work + modern AI tools = RepFlow.

Here's my stack:

Frontend:  React + Tailwind CSS → deployed on Vercel (repflow.cloud)
Backend:   FastAPI / Python → deployed on VPS (Hostinger)
Database:  MongoDB Atlas
Auth:      JWT + bcrypt

The Architecture Decision That Changed Everything: Multi-Tenancy

RepFlow isn't just for me — it's a SaaS for all commercial representatives. Every tenant needed complete data isolation from day one.

My solution: a tenant_id field on every single document in MongoDB.

# Every API endpoint filters by tenant
@router.get("/orders")
async def get_orders(current_user: User = Depends(get_current_user)):
    orders = await db.orders.find(
        {"tenant_id": current_user.tenant_id}
    ).to_list(length=None)
    return orders

Simple. Effective. Every query is automatically scoped to the tenant. No data leaks between clients.

If you're building a SaaS, architect for multi-tenancy from the first line of code. Adding it later is a nightmare.

What RepFlow Does Today

After months of building — mostly evenings and weekends — RepFlow has:

Client Management

• Customer health scoring
• Automatic address lookup via ZIP code
• Market segment tagging Operational Modules
• Orders with open balance tracking
• Full quote → order funnel
• Invoice management
• Commissions with bulk actions Reports & Analytics
• KPI dashboard
• Tonnage: portfolio vs. billed
• ABC Curve analysis
• Client activation tracking
• Revenue by period Infrastructure
• Multi-tenant architecture (tenant_id on everything)
• Super Admin panel
• In-app notifications
• PDF generation (orders, commissions, client reports)
• Security audit: 17 vulnerabilities identified and fixed

The Hardest Part: Security

When I first deployed, I ran a security audit. Found 17 vulnerabilities.

Things like:

• Missing rate limiting on auth endpoints
• JWT tokens with no expiration
• MongoDB injection possibilities
• Unprotected admin routes The 14-year-old who used to dig into how systems worked — he came in useful here. Each vulnerability I researched, understood, and fixed. This part wasn't glamorous. But it taught me more about real production systems than any tutorial ever could.

What I Learned

1. Domain knowledge is a superpower.
I didn't need to imagine what the software should do. I lived the problem for 15 years. That clarity accelerated everything — and no junior developer could replicate it.

2. AI doesn't replace understanding — it requires it.
If you don't understand what you're building, AI-assisted development produces working code that does the wrong thing. The curiosity to understand systems matters more than ever.

3. Ship early, fix constantly.
RepFlow went live before it was "ready." Real usage revealed problems no amount of planning would have caught.

4. Your past is not irrelevant.
Every macro I wrote in VBA, every HTML page I built at 14, every hour spent understanding how systems worked — it all compounded into this. Nothing was wasted.

5. Implementing software is different from using it.
I was part of a full ERP rollout covering accounting, inventory and sales. Watching real users adopt — and resist — a system taught me more about UX and product decisions than any design course. When I built RepFlow, I already knew which features would actually be used and which ones would be ignored.

What's Next

I'm currently building Cloudmetric — an IoT monitoring system for short-term rentals. ESP32 sensors reading temperature and humidity every 2 minutes, LG ThinQ AC integration, real-time dashboard, Eco Score and ESG reporting for Airbnb hosts across Europe.

I'm also moving to Lisbon to apply to 42 Lisboa — the peer-to-peer programming school with no teachers, no lectures, pure problem-solving. Feels like the right next chapter.

The Point

You don't need a CS degree to build real software.

But you do need something most tutorials won't give you: a real problem you understand deeply, the patience to learn how systems actually work, and the honesty to fix what's broken.

The curiosity that started at 14 — that's what got me here.

RepFlow is live at repflow.cloud.

Built with React · FastAPI · MongoDB · Claude · Cursor

Follow me here on dev.to — next article: building Cloudmetric, an IoT system for Airbnb hosts in Europe.