From Startup Idea to AI-Powered SaaS Product

The Problem

A Netherlands-based client had a startup idea: automate a digital marketing workflow — publishing ads to Google Ads — that currently took their team multiple days to complete manually. They needed someone to take this idea and turn it into a real, scalable SaaS product.

The challenge wasn't just building software. It was translating a non-technical founder's vision into a viable technical architecture, making critical decisions about AI integration (cost vs. quality), and leading a team to deliver a functional product from zero.

Constraints

• Startup stage — no existing codebase, no existing team processes, limited budget
• AI token costs could easily make the product economically unviable if not carefully managed
• The client had strong product vision but limited technical background — translation between business goals and engineering decisions was critical
• Freelance engagement — needed to balance architecture, hands-on coding, team leadership, and client communication simultaneously

My Role

Solutions Architect + Lead Engineer. I was responsible for the full technical direction: architecture decisions, team leadership, development execution, delivery planning, and client communication. I worked across all layers — from system design to shipping features.

The Architecture

The platform used Symfony on the backend with a React/Next.js frontend, deployed across Vercel (frontend) and Scalingo (backend API). The AI integration was the core value proposition — the system needed to generate marketing content at scale while keeping costs sustainable.

Key Decisions

• AI token optimization: The hardest decision was the tradeoff between AI response quality and token cost. Raw, unoptimized usage would have burned through budget and made the product unprofitable. I found a middle path: restructured prompts and workflows to reduce token consumption by approximately 50% while maintaining response accuracy. This wasn't just a cost saving — it made the product economically viable as a SaaS.
• Symfony + Next.js split: Chose Symfony for the backend (reliability, my team's strength, complex business logic) and Next.js for the frontend (SSR for marketing pages, fast iteration for the UI). Vercel for frontend, Scalingo for backend — each deployed to its optimal platform.

The Outcome

• Transformed a multi-day manual workflow into a process that completes in hours
• Reduced AI token consumption by ~50% without sacrificing output quality — making the SaaS unit economics viable
• Moved the product from concept to functional SaaS platform
• Delivered in a freelance context while managing architecture, engineering, team, and client expectations simultaneously

What I'd Do Differently

I'd push harder for a proof-of-concept phase before committing to the full architecture. We made some early decisions based on assumptions about AI behavior that we could have validated faster with a focused POC. I've since adopted this as my default approach — for clients with technical background, I now lead with RFCs and POCs before full implementation.