Taking a vibe-coded tutoring marketplace to production

A tutoring marketplace a founder built in an AI app builder over two weekends: it worked in a demo and fell over with real parents paying. The concept is the six weeks it would take to make it safe to launch without starting again.

Concept project. Designed by OnTek to show how we would solve this problem; not work done for a client. The figures are the targets it was built against.

Product: Tutorloop
Industry: Education
Year: 2026
Timeline: 6 weeks

What it is built to do

The targets the build was designed and tested against. A concept, so these are design targets rather than a client’s measured results.

6
Weeks, not a rewrite
The plan from audit to launch, keeping the founder’s product.
0
Open data paths
Every table locked to its owner, checked by tests on each release.
1
Charge per booking
Payments made idempotent, so a retry never bills a parent twice.

Thebrief

The prototype matched parents with tutors, booked sessions and took payment. It also kept every user’s data readable by every other user, charged some bookings twice, and had no tests, so every fix broke something else.

The founder did not want a rewrite. They wanted the product they had built, made safe enough to take real money from real parents.

How we ran it

Each step was owned by one discipline, and every one of them worked from the same plan in the same team.

01
Vibe Coding Assistance
An audit in plain language
A week reading the generated code and writing down what would hurt first: open data access, payments without idempotency, secrets in the client. Ranked by risk, each with a fix and an estimate.
02
SaaS, MVP Development
Fixing it in place
Row-level security on every table, payments that cannot charge twice, and the tangled booking logic pulled into one place a future hire can read, keeping the founder’s screens as they were.
03
SQA, Deployment
A safety net before launch
Tests around booking and payment, a staging copy of the app, and a release pipeline, so the founder can keep building with AI tools without breaking what now works.