MedRank

MedRank pulls together the two things that actually tell you whether a hospital is a good place to train (CQC inspection ratings and GMC National Training Survey scores) for over 500 NHS hospitals, and puts them in one searchable place.

It's aimed at the people who have to rank training jobs with almost nothing to go on: foundation doctors picking specialty posts, core surgical trainees ordering their preferences, anyone staring at the Oriel spreadsheet wondering how they're meant to choose.

That last group is what Oriel Ranker is for, and it went live this year. You upload the Oriel export exactly as it comes, and MedRank reads it, matches every placement to CQC and GMC data, and lets you rank the whole list against quality and travel instead of guessing. A core surgical offer list can run to 649 rows of free text. That is the problem the app now solves end to end.

Four things now, and the fourth took the longest.

Preferencing works like this: you are sent a spreadsheet of offers, written in whatever format your region felt like using, and asked to put it in order. Free-text placement descriptions, trust names spelled four different ways, hundreds of rows. Everyone does it by hand, badly, in a weekend.

Oriel Ranker takes the export as-is. It parses the rows, matches the hospitals to the CQC and GMC data already in MedRank, and gives you a ranking screen where every rotation carries its quality scores and travel burden next to it. You reorder until it looks right, then export the order back to Oriel rather than dragging every post by hand.

Parsing that mess reliably is the whole engineering problem, and most of this year's work went into it: large files that time out, retries that used to lose your saved rotations, uploads superseded halfway through. Pricing follows the same logic. Ranking, reordering and exporting are free. You pay £14.99 once, at the point you upload, because the upload is the expensive part. If it fails, the credit goes back.

When you're offered training posts, you rank your preferences with barely any real information. Most people go off reputation, a forum thread from 2019, or whatever a friend of a friend said.

The better data was always there. CQC and GMC both publish it. But it lives on two separate sites in two separate formats, and cross-referencing it by hand is enough of a pain that almost nobody does it.

I'd gone through preferencing myself and felt exactly this. It seemed daft that the thing deciding three years of my life came down to vibes. So I built the tool I'd wanted.

Next.js and PostgreSQL with Prisma, Mapbox for the maps, Tailwind for styling, hosted on Vercel and pinned to London so uploads run near the database. Oriel Ranker added the parts a read-only site never needs: accounts through NextAuth, payments through Stripe, and an AI parsing pass that turns free-text Oriel rows into structured placements.

The harder call was the data, not the framework. Neither CQC nor GMC offers an API, so MedRank runs on its own database that I refresh rather than pull live. That's fine. A trust changes its EPR maybe once a decade and GMC data updates once a year. The honest move was a clear "last updated" date and a link back to the source, not pretending it's real-time.

I built the first version on Lovable in a weekend. The frontend looked great. Then I tried to connect a backend, couldn't, read documentation I didn't understand, and eventually closed the tab. I didn't open it again for two months. I'd quietly decided I'd been kidding myself.

When I came back, I started over in the opposite order: database first, then the APIs, then the interface. That single change is the only reason MedRank exists. I wrote about the two months I quit in more detail on the blog; it's the part of these stories people usually leave out.

The other lesson was about me, not the code. I added the Digital Maturity filter because it was interesting to build, not because anyone asked for it. That's a pattern I've had to learn to watch. MedRank is live and does its job. The harder discipline was knowing when to stop adding to it.