The Problem

Public procurement in Slovakia is spread across multiple disconnected portals like UVO, TED, Vestník SR, EKS, Josephine, and Úradná Nástenka. For SMBs and consultancies, that usually means spending hours every day filtering irrelevant notices just to find a few tenders that actually fit their business.

I built TenderMonitor.sk to simplify that process completely. The goal was to create a system where users no longer need to manually search procurement websites all day, but instead receive a feed that already understands what is relevant for them.

The Product

TenderMonitor.sk is a Slovak-first SaaS platform that aggregates procurement data from all major Slovak and EU tender portals into one clean interface.

The platform runs every tender through a multi-stage AI matching pipeline combining negative keyword filtering, CPV matching, semantic stem analysis, and final Gemini scoring based on each user’s own relevance threshold.

Instead of relying on simple keyword alerts, the system uses Google Gemini Flash models to better understand the context, intent, and overall “vibe” of a business and match it with contracts that actually make sense for them.

Users onboard with their keywords and CPV codes, and after setup they receive a highly curated feed with almost no noise. On top of that, the platform sends automated daily digests through Resend with new matches, AI scores, and short reasoning summaries so important opportunities never get missed.

Design Engineering

TenderMonitor.sk was a full product build where I combined product thinking, UX, frontend engineering, and AI-driven workflows into one system.

I designed and built the whole experience end-to-end, from the information architecture, Slovak microcopy, onboarding flows, and visual identity, all the way to frontend implementation, background workflows, and operational tooling.

A big focus for me was making a technically complex product feel simple and reliable. The platform processes large amounts of procurement data and AI scoring in the background, but the user experience stays lightweight, fast, and easy to understand. Things like optimistic Stripe flows, real-time state updates, and automated daily digests were designed to make the product feel responsive and alive instead of feeling like a traditional enterprise procurement tool.

I also approached the project from a systems design perspective, not only as a UI layer. Features like GDPR-compliant deletion, session management, notification preferences, scraper monitoring, and AI-assisted repair workflows were designed as connected parts of the overall product ecosystem from the beginning.

On the internal side, I built the Scraper Pulse dashboard as an operational control center for monitoring scraper health, debugging failures, reviewing logs, and approving AI-generated fixes in real time.

The Most Interesting Part: Self-Healing Scrapers

The biggest challenge with procurement aggregation is not collecting the data initially, but keeping scrapers alive when government portals suddenly change layouts without warning.

To solve that, I designed a self-healing recovery loop that continuously monitors scraper stability and proposes automated fixes.

The process works like this:

  • Every scraper compares the current page structure against a stored structural fingerprint during each run.
  • If something breaks, the system automatically captures screenshots and HTML diffs using Puppeteer and Chromium.
  • That evidence is analyzed by a dedicated Gemini agent that compares the old and new layouts.
  • The AI then proposes a potential patch, usually updated selectors or parsing logic, directly inside the Pulse dashboard for one-click approval.

Instead of silently failing, the system can detect problems, investigate them, and suggest its own repairs before the pipeline fully breaks.

AI-Assisted Workflow

The platform itself was built through a custom AI-assisted workflow powered by around 20 specialized agents covering frontend, backend, debugging, security, SEO, QA, and deployment tasks, along with dozens of reusable skill modules and structured workflows.

My role was defining the product vision, UX direction, architecture, and design system, while also guiding the workflows, reviewing outputs, and making the final product decisions. The agents handled execution based on tightly scoped specifications and verification flows before anything shipped.