I Built This Entire Platform in 48 Hours (Here's How)

Friday morning: domain purchased. Sunday evening: live in production with paying customers.

This is the complete build log of shipkit.io. Every decision. Every prompt. Every fuck-up. Complete transparency because that's how you learn.

Hour 0-2: Planning and Architecture

Most people skip this. They jump straight into code and end up rebuilding everything three times. I spent two hours mapping out:

• Core value prop: AI builds your software, humans review and ship it
• User flow: Upload voice note → AI generates proposal → Client approves → We build → Deploy
• Revenue model: Three tiers (Validation $47, Launch $197, Turnkey $497)
• Tech stack: Next.js 14, Prisma, Supabase, Anthropic Claude, Stripe

Hour 2-8: Database Schema and Auth

This is where AI saves the most time. Setting up authentication used to take me a full day. Now?

"Set up NextAuth.js with email magic links, Google OAuth, Prisma adapter, and session management. Include user roles and permissions system."

Boom. Done in 45 minutes. I reviewed the code, tested edge cases, deployed.

The Prisma schema was similar. I described the data model:

• Users with authentication
• Projects with workflow states
• Proposals with pricing tiers
• Tasks with dependencies
• Deliverables with artifact storage

Claude generated the entire schema. Relationships, indexes, constraints—all production-ready.

Hour 8-16: The AI Agent System

This was the hardest part. Building an orchestrator that manages multiple AI agents, each with different specializations. Here's what I built:

I built the base agent class first. Then each specialist inherited from it. The orchestrator manages the queue and passes context between agents.

Hour 16-24: Client Portal and Dashboard

The client-facing UI had to be dead simple. No learning curve. Upload voice → See proposal → Approve → Track progress.

I used AI to generate the entire dashboard:

• Project cards with real-time status
• Proposal viewer with approve/revise actions
• Progress tracker with phase breakdowns
• Deliverables download area

The cyberpunk theme (pink/cyan/green gradients on black) came from wanting something that stands out. Every other dev agency site looks the same—blue gradients, generic tech imagery. This needed personality.

Hour 24-36: Payment Integration and Billing

Stripe integration. The thing that used to take me 2 days. AI prompt:

"Integrate Stripe Checkout with three pricing tiers. Handle webhook events for successful payments, failed payments, and subscription management. Update project status in database after payment confirmation."

Two hours later: complete payment system. Webhooks working. Database updates triggering. Email confirmations sending.

I tested with Stripe test mode, caught a race condition in the webhook handler, fixed it, pushed to prod.

Hour 36-42: Logo Generation and Branding

I integrated Dumpling AI (FLUX model) to auto-generate logos for client projects. Each project gets 5 logo variations based on their brand strategy.

The flow: Client approves proposal → Orchestrator triggers task decomposition → Brand strategy agent creates visual identity → Dumpling generates logos → Client gets ZIP with all assets.

This was pure value-add. Clients don't expect logo generation. But when they see 5 professional logo options included in their package? Instant wow factor.

Hour 42-48: Testing, Debugging, Deployment

The final stretch. Testing every user flow:

• Sign up → Upload voice note → View proposal → Approve → Track project
• Payment failures and retry logic
• Edge cases (empty voice notes, unclear requirements)
• Mobile responsiveness
• Accessibility (WCAG 2.1 AA compliance)

Found and fixed bugs:

1. Voice transcription failing on long pauses (added silence detection)
2. Proposal generation timing out on complex projects (added streaming)
3. Mobile menu not closing on navigation (fixed state management)
4. Database connection pool exhaustion (increased pool size, added timeout)

Deployed to Vercel. Connected custom domain. Set up monitoring with Sentry. Live in production 48 hours after starting.

What I'd Do Differently

Honestly? Not much. The architecture held up. The AI agents work. The user flow is smooth.

But if I were starting over:

• Plan context management first. I had to refactor this twice. Should've nailed it upfront.
• Set up monitoring earlier. I deployed without Sentry. Bad idea. Caught production bugs by accident.
• Write better prompts initially. I got lazy with some prompts, got mediocre code, had to iterate.

The Reality Check

Could I have built this without AI? Yes. Would it have taken 2-3 weeks? Absolutely.

Authentication, database design, API routes, agent orchestration, payment integration, UI components, mobile responsiveness—all of it would've been manual. All of it would've accumulated bugs at 11 PM when I'm tired and cutting corners.

AI doesn't get tired. AI doesn't skip error handling. AI generates production-ready code if you give it good prompts and review the output.

Want the Prompts?

I'm documenting every prompt I used to build this. Complete engineering logs. What worked, what didn't, what I'd change.

Building in public means showing the whole process. Including the messy parts.

More posts coming soon. Subscribe or bookmark this page.