Pixicular

Problem

Businesses that need image analysis — age verification, gender detection, object and facial detection — generally don’t want to run their own machine learning infrastructure. They want a reliable API, a straightforward billing relationship, and a dashboard to check what was analysed and when. That gap is what Pixicular was built to fill.

Approach

We built and launched Pixicular end-to-end as a B2B SaaS product: a customer-facing web app for signing up, managing a plan and reviewing analysis history, sitting in front of a REST API doing the actual ML work. Launching it ourselves — not just building it to someone else’s spec — meant making the same production trade-offs (pricing tiers, rate limits, infrastructure cost) that any real SaaS operator has to make.

Stack

A Fastify (Node.js) REST API providing multiple ML-powered analysis endpoints via Amazon Rekognition, with a Next.js frontend authenticated through Clerk. Billing runs on Stripe Payments and Subscriptions across tiered plans. Infrastructure is a Hetzner-plus-AWS setup (S3, CloudFront, Route 53), entirely managed through Terraform. A custom image-classification model built with Hugging Face is currently in active development, to extend what the platform can detect without depending solely on third-party recognition services.

Outcome

Pixicular is live in production with real paying customers on tiered Stripe subscriptions — a full SaaS stack (auth, billing, infrastructure-as-code, ML-backed API) that we designed, built and continue to operate ourselves. It’s also the clearest single piece of evidence for four of our services at once: the API is our backend work, the customer app is our frontend work, the Terraform-managed hosting is our cloud & infrastructure work, and the Rekognition and Hugging Face integration is our AI & automation work.