Ionut-Cristian Florescu

AJM Fleet Purchase buys new and used cars in bulk across Europe and sells them to fleet operators in Europe, Asia, the Middle East and Africa. From 2020 to 2023, as a contractor, I built their back office from a blank repo and ran it in production.

It covers the whole trade, from the purchase of each car to the last payment. Every vehicle is followed from supplier to buyer, with its VIN, estimated and actual prices, fees and delivery dates. Proforma and final invoices are issued for several group companies in several currencies, generated as PDFs and sent from the app. Incoming and outgoing bank payments are matched to vehicles and invoices, with a balance sheet per vehicle exported to Excel. A dashboard shows deliveries due, overdue invoices and records that don't add up.

I wrote it alone, in more than 1,100 commits, all before AI coding assistants existed.

LeasingSH.ro has been live since 2010 as Romania's first online marketplace for new and used cars, in leasing or for cash. I joined in 2013 as technical partner and fractional CTO, in a profit-share partnership, and I've owned the technical side end to end ever since: architecture, code, infrastructure, data and operations.

In those thirteen years I've rebuilt it three times (2013, 2015 and 2025) as the web changed, and the business kept running through each migration. Today it lists over 30,000 cars from more than 750 dealers in Romania and Germany, updated daily.

Behind the listings is a data pipeline. It ingests offers from the main Romanian listing sites, a dealer API and an assisted importer, then uses LLMs to enrich and clean up the data, so that thousands of inconsistent listings become one searchable catalog. Around that sit lead tracking across calls, WhatsApp and messages, and an admin console and editorial CMS for dealers and partners.

It's also ready for AI assistants. A public API with MCP and OpenAPI descriptors lets Claude, ChatGPT and other assistants search the offers directly.

PocketBase packs a whole backend into a single file, but as a project grows, its collections and the relations between them get hard to keep in your head. I built PocketBaseUML in March 2023 to make them visible: a free, open-source tool that draws a UML diagram of a PocketBase database.

You can connect it to a live server with your superuser account, or paste the collections JSON exported from PocketBase's admin. Either way, you get a diagram of every collection, its fields and its relations, which you can zoom and download as a PNG. It's a static web app that runs entirely in your browser, with no backend of its own. Use it online at pocketbase-uml.github.io, or run it locally with npx pocketbase-uml to reach a server on your own machine. It handles both current PocketBase exports and those from before v0.23, so older projects still work.

It's a small, focused tool, and PocketBase developers have given it 160 GitHub stars. I still maintain it: the latest release came out this month.

In February 2022, SvelteKit was young and tRPC was quickly becoming the way to share types between server and client in TypeScript apps, but there was no clean way to use the two together. I wrote tRPC-SvelteKit to fill that gap, and its tagline has stayed the same since: move fast and break nothing.

The idea is simple. You define your tRPC router on the server, add a single handle hook, and call your procedures from your pages with full type safety from the database to the UI. It works with every SvelteKit adapter: Node, Bun, Vercel, Netlify, Cloudflare and more. Procedures can set cookies and headers, responses can stream, subscriptions run over server-sent events, and there's experimental WebSocket support. Since today's 4.1 release, it also generates typed TanStack Svelte Query options.

I've kept it current for almost five years: three major versions of tRPC (v9 to v11) and three of SvelteKit, with 119 releases, 20 contributors and an upgrade guide for every breaking change. It has a dedicated docs site, which now publishes llms.txt so AI coding assistants can read the docs too.…

Mantine DataTable is the table component for data-rich React apps built with Mantine. It handles what a real back office needs: asynchronous data loading, pagination, sorting, row selection, row expansion, context menus and dark-theme support, all fully typed, lightweight and dependency-free. Teams at Microsoft and Namecheap use it, along with kapa.ai, InvenTree and many others.

It gets over 750,000 npm downloads a month.
I've built and maintained it since 2022, pre-AI, with full docs and a legacy version still supported for teams that can't upgrade yet.