Shipping Faultrix: What I Learned Building an AI SaaS in 5 Months
What changed when I moved from research into product building: Faultrix, an AI SaaS for construction quality control that generates ONORM-aligned reports in under a minute.
From Research to Product
Faultrix is an AI-powered construction quality-control SaaS. A user uploads site photos, the system analyzes them, and the platform generates an ONORM-aligned report in under a minute.
Building it taught me that the jump from research to product is not mostly about choosing another model. It is about making the whole system useful.
The Product Stack
Faultrix was built with:
- Next.js
- Convex
- OpenAI API
- Clerk
- Cloudflare R2
- Stripe
- Docker
The stack was chosen for speed of iteration and operational clarity. I wanted to spend time on product flow, reliability, and reporting quality, not on boilerplate infrastructure.
The Hardest Part Was Not the Model
The AI side was important, but it was not the hardest part. The hardest part was product fit:
- shaping outputs so they matched real reporting expectations
- keeping the user flow short and clear
- handling evidence and storage in a way that felt trustworthy
That is where research instincts help and fail at the same time. Rigor transfers well. Product intuition has to be earned in the field.
What Research Helped With
- structured experimentation
- testing output quality before shipping
- understanding where the model should ask for human review
What Research Did Not Automatically Teach Me
- pricing
- friction in onboarding
- how fast users lose patience
- how much UX can matter more than raw model capability
The Main Lesson
The first version of Faultrix had strong AI and too much workflow friction. That taught me the key product lesson: users experience value through the path, not through the architecture diagram.
If the path to value is too long, the sophistication underneath barely matters.
Frequently Asked Questions
What is Faultrix?
Faultrix is an AI-powered construction quality-control SaaS. Users upload site photos and receive a structured ONORM-aligned report with evidence handling and security-minded storage.
What stack powers it?
Next.js, Convex, OpenAI API, Clerk, Cloudflare R2, Stripe, Docker, and a Python-heavy AI workflow behind the scenes.
What was hardest?
Not the AI itself. The hardest part was making the product fit real workflows, legal formatting expectations, and a low-friction user experience.