One desk for a whole tutoring operation.
An integrated operations platform for Starline Scholars: a Windows desk for the founders, a managed backend, parent onboarding and payments, tutor payouts and a read-only AI agent — all working from one operational record.
Last updated
- Partner
- Starline Scholars
- Sector
- Tutoring
- Capabilities
- Software, data and AI
- Delivered
- Desktop app, hosted system, payments, AI agent
- Status
- In production
At a glance.
From the overview through settings.
Run on every change, over the rules that cost money to get wrong.
A morning sweep and an evening reminder.
Alina, read-only, answering from the live record.
Maths tutoring
GCSE and A-Level
32-hour courses
Paid once, up front
The partnership.
Starline Scholars is a maths tutoring business for GCSE and A-Level pupils. A pupil holds two weekly evening hours with a tutor and works through a 32-hour course: two hours a week for sixteen weeks. Parents pay once per course, up front. There is no subscription and no invoice to chase later; a pupil who carries on renews with another single payment.
That is simple to describe and unforgiving to run. Hours have to be matched to tutor availability and subject. A lesson nobody turns up to still uses up an hour; one called off in advance does not. A pupil leaving has to free their hours, cancel what was scheduled and tell the parent. Tutors have to be paid for what was actually delivered, not what was planned, and enough money has to stay in the account to pay them. All of it lands on the founders’ week.
AnotherAgent is Starline Scholars’ technology partner: one partner across software, data and AI, from the first build through to keeping it running.
Six parts
One definition
What we built.
The desk the founders work in, the rules behind it, the pages parents and tutors see, the money in and out, an AI agent on the desk, and one shared definition underneath all of it.
Starline Operations Desk
A Windows application the founders open each day. Overview, onboarding, hour changes, parents, pupils, tutors, the timetable, sessions, money, the financial account, the mail log and settings sit on twelve screens, all reading from the same record. It installs from one download and updates itself.
The hosted system behind it
Every rule the business runs on lives server-side rather than on the desktop, so a change takes effect everywhere at once. Two scheduled runs a day keep it moving: in the morning it books the next lessons, settles tutor pay, chases any payment still open and checks every refund against Stripe; in the evening it reminds each parent and tutor about tomorrow’s lessons.
Pages for parents and tutors
One link takes a parent through the payment, their child’s details and choosing two weekly hours with a tutor. Later links let them move an hour or renew for another course. Tutors join from an invite link, set their subjects and evenings, and give their bank details to Stripe directly.
Money in and money out
Parents pay on a page Stripe hosts, so card details never reach our system. A parent who paid and never added their child can be refunded in full from the desk. Tutors are paid into their own bank accounts through Stripe. The Money page shows what came in against what went out over twelve months, and holds back what tutors are owed, plus a week of cover, before anything can be withdrawn. Stripe’s balances sit on a separate financial account screen.
An AI agent on the desk
Alina, a chat window on every screen, so the founders can ask how to do something, look someone up or ask what the live record says. It reads and never writes, and it stays usable while a pupil or tutor is open on the desk. It runs on Gemini 2.5 Flash through the Vercel AI Gateway, with a smaller OpenAI model if that is unavailable, and each answer shows how long it took and what it cost.
One shared definition of the domain
The hour grid, seat occupancy, tutor supply, pay rules and course progress are defined once and used by both the desk and the server, so the two can never disagree about what a seat is or what a tutor is owed.
The whole operation, on one desk.
A pupil holds two hours a week until their 32 hours are delivered, then completes and frees the seat. A recording of the desk itself, running on sample data, from the overview through the timetable, onboarding, the tutors, the week’s sessions and the money.
A recording of the desk, running on sample data. The pupils and tutors shown are invented; the subjects, year groups, evening hours and 32-hour course are the real model the system runs on.
The decisions that matter.
This system holds children’s names and parents’ contact details, and it moves money to tutors. That shaped the build more than any feature did.
The desktop app holds nothing
No local database, no payment handling, no forms of its own. The desk is a window onto the hosted system, which means a lost or stolen laptop is an inconvenience rather than a data breach.
Personal data is protected, then deleted on time
Pupil and parent details are encrypted in the database, sign-in links are never stored in a readable form, and the founders’ accounts support a second factor at login. A daily job deletes records on the schedule the privacy policy sets, and a test holds the two to the same numbers.
Nobody who has paid is left looking unpaid
A completed payment is reconciled against the right child without anyone checking by hand. If Stripe’s confirmation never arrives, the system asks Stripe directly when the parent returns, and the morning run asks again about every payment still open and every refund still unanswered.
Tutor bank details never reach the system
Tutors give their bank details to Stripe on Stripe’s own page; the desk only ever sees the last four digits. Pay is held back before anything can be withdrawn, so the money for tutors is there when it is due.
The AI agent can read, never write
It answers from a guide checked against the desk and from the live record, through a database role that can only read. The database itself withholds passwords, sign-in links and payment identifiers from it, and a hard daily spending cap limits what it can cost.
Tested where it would be expensive to be wrong
More than 880 automated tests cover seat allocation, course hours, tutor pay, payouts, refunds, reminders, retention, the AI agent’s access and the access rules. They run on every change, and the whole money path is exercised end to end against a stand-in for Stripe.
Frequently asked questions
What did AnotherAgent build for Starline Scholars?
A Windows operations desk for the founders, the hosted system and database behind it, the pages a parent uses to enrol a child, pay and pick their weekly hours, the pages a tutor uses to join and get paid, and a read-only AI agent inside the desk. Overview, onboarding, the roster, the timetable, sessions, reminders, money, the financial account and the mail log all run in one place rather than across separate files.
How does the platform keep every workflow consistent?
The desk, the parent and tutor pages and the payment flow all use the same hosted domain model. Seat availability, course hours, tutor pay and payment status are defined once, so every part of the platform works from the same operational record.
Where does pupil and parent data live?
On hosted infrastructure we manage, behind a login, never listed or indexed. Personal details are encrypted at rest, sign-in links are never stored in a readable form, and records are deleted on the schedule the privacy policy sets.
Does the Windows app hold the data?
No. The desk has no local database and handles no payments — it is a window onto the hosted system. A lost or stolen laptop is an inconvenience rather than a data breach.
What does the AI agent on the desk do?
Alina answers the founders’ questions inside the desk: how to do something, where to find a pupil or tutor, and what the live record says. It is read-only, so it points to the right page and button rather than changing anything, and the database itself withholds passwords, sign-in links and payment identifiers from it. It runs on Gemini 2.5 Flash through the Vercel AI Gateway, with a smaller OpenAI model behind it, under a hard daily spending cap.
How are tutors paid?
For the hours they actually delivered, straight into their own bank account through Stripe. A tutor enters their bank details on Stripe’s own page, so those details never reach Starline’s system, and the desk holds back what tutors are owed, plus a week of cover, before the founders can withdraw anything.
Who owns the software at the end?
Starline Scholars does. The accounts, the documentation and the running product are theirs. We stay because they want us to, not because the system cannot run without us.
Tell us the job you would take off your plate first.
Send a few details and we will come back with whether it is a strong fit and what it would take. No obligation, no jargon.