Evgeny

UVAGA pulls together Portugal's public healthcare data, today scattered across hundreds of Excel files, CSV exports, Power BI dashboards, and government portals. Every source has its own format, structure, and naming. One registry uses NIF codes, the next doesn't. The same clinic shows up under three different names. Getting all of that to line up was most of the work.

Collecting it was only half the job. I cleaned and reshaped everything, then built each workspace around the group using it. A hospital group, a diagnostics operator, an insurer each comes at the data with different questions. Cut it the right way and patterns show up that were buried in the raw files.

It didn't start as a B2B product. The first idea was simple and human. Uvaga.Emergency helps people find the fastest emergency room by real wait times and distance, and Uvaga.Consultations does the same for clinics and appointments. Both were built to take some pressure off a health system that's always overloaded.

Then I noticed the same data was just as useful to businesses, not only to regular people. A bit of research turned up several distinct groups who needed it, each asking different questions. So I rebuilt UVAGA around them and turned it into a B2B platform.

It started as a weekend pet project and grew into a platform I took from zero to one on my own.

I leaned heavily on deep research and sub-agents across the market to find who actually had this pain and what it cost them. The audience split into four groups: investors, hospital groups, diagnostics and labs, insurers. They all needed the same underlying data, just cut a different way. So each group became its own workspace, with dashboards built for how that group thinks.

From there I worked through the scenarios for each group: how they really work with the data, what questions they bring, what they actually need to see. That turned into a customer journey map and a user flow per group, and the product is built around those instead of dumping charts on a page. The investor workspace, for example, follows a funnel: market thesis → sector fragmentation → region → targets → deal model. Every group got its own flow like that.

Portugal's healthcare data is public, but public doesn't mean usable. It sits trapped inside portals, dashboards, PDFs, and spreadsheets, none of it meant to be pulled out in bulk. Every source is its own little world, with its own format, structure, and quirks, scattered all over the place. I wrote a scraper for each one. The hard part wasn't grabbing the data, it was getting all of it to agree: cleaning it, lining it up, and stitching it into a single model that actually holds together.

23 sources in all, cleaned and stitched into one model: 39,713 facilities, 742,150 datapoints, ownership resolved by tax ID so you can see who owns what. A pipeline refreshes everything on a schedule and runs dozens of quality checks a day, so a broken source upstream doesn't quietly poison a number downstream.

With the data in place, I designed and built the product itself. It's a monorepo: the consumer site, the B2B platform, and a shared API that feeds both.

The front end is React, TypeScript and Vite, with Mantine for the UI and charts, TanStack Query and Zustand for state, and Leaflet, Turf.js and OpenRouteService for the maps. That stack powers the choropleths, the 30,000+ clustered provider pins, and the drive-time catchment areas. The whole interface runs in three languages: English, Portuguese and Spanish.

The back end is a Fastify API on Node with Prisma and Kysely over PostgreSQL: 135 endpoints feeding 36 dashboards across the workspaces. Behind it sits session-based auth and multi-tenancy, so each organization sees only the workspaces and tier it pays for. Scrapers run on Playwright.

The whole thing runs on a single server I manage myself, behind Nginx with PM2, SSL, and daily cron jobs. A Telegram bot pings me the moment something breaks. A CI pipeline blocks any change that breaks the math, and 761 unit tests run again every night.

A product nobody hears about doesn't exist, so growth was on me too. The whole marketing side, solo, with the same AI and sub-agents I used to build the product.

I built the public landing page at Uvaga.Insights, wrote and published the articles on LinkedIn that funnel readers to it, and designed every visual along the way, banners included. I drew up targeted lists of the private-equity firms and operators who'd actually care, ran the outreach myself, and talked to the people who replied. Then I tracked the numbers: who visits, what they open, what converts, and tuned from there.

Positioning, copy, design, distribution. Same person.

I work solo, so AI is how I cover four jobs at once. Most of it runs through Claude Code in VS Code: planning, analysis, documentation, and the bulk of the design and engineering, with sub-agents working in parallel. Perplexity handles the market and UX research. ChatGPT and Gemini help with content and UX writing. Leonardo, Midjourney and Nano Banana make the visuals and video. Lovable and Figma AI are for quick prototypes.

None of it replaces judgment. It just closes the gap between deciding something and shipping it.

A complete B2B platform, taken from a blank page to a product people log into and use.

Idea, research, data, design, code, and the selling. One person, end to end. That's the real point here: with the right AI workflow, a single person can own an entire platform, 0 to 1.