SchoolCost

My cousin's university search started it.

My younger cousin was heading into SS3, her final year of secondary school. I sat down to find good Nigerian universities she could go to and to put a figure on each one. A lot of the costs, and almost all of the breakdowns, weren't anywhere I could find. When I later went school by school, I saw why: old PDFs, portals that wanted a student login, and blogs that each quoted a different amount.

Around then, I finished Dára Sobaloju's Ship With AI cohort. Afolabi Sokeye's Salary Index gave me a starting point: one public table anyone could check. I began building the table I couldn't find and posted about it on LinkedIn two days later.

The question I kept hearing. Under that first post, Faari Joye asked: "how'd you get the data on each school? Is there an API you used?" There isn't one. That question shaped the whole product.

The fee isn't the cost, and nobody agrees on the fee.

A family choosing a university usually compares one number per school: the tuition. The real cost is spread across acceptance fees, departmental and ICT charges, medical fees and the hostel, some paid once and some every year. And the amount for a first-year student isn't the amount for a final-year one.

Finding even the tuition is hard. Some schools publish an itemised PDF on their own site. Others hide it behind a student portal, mention it in a news post, or don't publish it at all. The gap gets filled by blogs, which copy each other and rarely say where a number came from.

Every number says where it came from.

To decide where to look, I sorted every school's sources into four tiers, and only the first counts as official.

A fee belongs to a course, a level and a student.

A family's first step is search. People type the way they talk, so it forgives small typos ("unilg" still finds the University of Lagos) and matches courses and states as well as school names. From there they land on a school, then a course, and that's where the price lives.

I checked that bug against Augustine University's own 2026/2027 fee schedule. The seed data was right; the page was reading the wrong column. Try it below.

Compare the same course, at the same level, in the same session.

The people who pay the fees see them first.

There are two ways in. When a search finds nothing, the page offers "Suggest a school" instead of a dead end, and the fee form has the same link under its school picker. Anyone who knows a fee can share it, with the document that backs it up.

What I like most is what people send. Four people have used the forms so far. Three suggested schools I hadn't listed: University on the Niger, Miva Open University and Kwara State University. The fourth sent the 2025/2026 Computer Science fee for the University of Ibadan's Postgraduate College, and didn't stop at a number. They attached the invoice from the student portal, every charge itemised, adding up to the ₦246,700 they entered. That's the habit the whole product is built on: a number comes with its proof. All four are still waiting for my review. In the screenshots below I've blurred names, emails, invoice numbers and dates.

Submissions have a rate limit, and the database has row-level security. Each record keeps the date it was last reviewed, which is what the page shows next to every fee.

I designed it in Figma, then directed AI agents to build it in two weeks.

I designed the two screens everything hangs on first: a homepage built around one search box, and a fee breakdown that lists every charge with when it's due and keeps accommodation out of the total. The trust labels you see live, "Official" or "Community" on every fee and a Verification tab on every record, came during the build. I added them with the agents once Faari's question from chapter 1 made it clear: people care where a number came from as much as the number.

Then I built it end to end with Claude Code, Cursor and Codex: Next.js, a Supabase database and PostHog analytics, deployed on Vercel. 105 commits between 16 and 31 July. The agents wrote most of the code. I set the rules they worked to, and they're written in the project's own files:

1. No deploying without me. Pushing to the main branch puts the site live, so no agent could push unless I said so in that conversation. "Commit" never meant "deploy".
2. Never weaken the trust rule. The project notes head it "do not weaken this", and one file in the code holds it, so no screen can restate it differently: unchecked or low-confidence fees can be shown, but never look verified or win a comparison.
3. Sort every source before using it. Tier A is the school's own current, itemised schedule. Tier D is aggregator blogs, which never get published as verified.
4. Plain copy. No em dashes in any copy or UI text, set as a rule Cursor always applies.

The QA pass that mattered. Just before launch I ran a full check and found two bugs that showed families the wrong money: 720 records across 25 schools were labelled "verified" when they weren't, and every level showed the Year 1 fee, the bug you tried in chapter 4. I also replaced placeholder comparisons with real schools offering the same course. The fee figures had to be right before launch.

2,586 fee records, each with its source.

SchoolCost has been live since the end of July 2026. It lists 149 schools, with published fees for 42 of them and fees good enough to compare on price for 31. Every record shows where it came from and when it was last checked.

Usage, 31 July to 20 October 2026. PostHog counted 243 visitors, not including my own test accounts. They opened 164 school pages and 112 fee breakdowns, and 49 of them searched the directory 333 times. Most of that came in launch week and August. 129 of those searches found nothing, like "heritage" in chapter 6, often a course or school I hadn't priced yet. That's the clearest list I have of what to add next. Compare is the gap I described in chapter 5: I can't yet say how often people use it.

Search has started to send people too. On 26 October 2026 Google Search Console told me schoolcost.ng had reached 40 clicks from Google Search in the past 28 days.

People noticed. My three LinkedIn posts about building and launching it drew 274 reactions and 22 comments. People asked where the data comes from, and one asked for schools abroad too.

What I'd change if I started again.

1. Talk to five families before building. I built from my cousin's search and the comments under my LinkedIn posts, and never sat with a student or parent. 129 of 333 searches found nothing, things like "federal universities" or a course I hadn't priced. Five conversations in the first week would have shown me what people look for before I chose what to price.
2. Launch with fewer schools, fully priced. 149 schools are listed, but only 31 can be compared on price, so many visits land on a school without a usable fee. I'd launch with fewer schools, every one priced in full, and add more as the data is checked.
3. Design the reviewer's side as carefully as the sender's. I put real care into the forms people use to send fees, but not into the routine behind them. Four submissions came in, one with an invoice as proof, and all four are still waiting. Next time I'd decide who reviews, how fast, and how the sender hears back before opening the forms.