We built NabavkiData to automatically analyze 15,000+ government tenders in North Macedonia using 50+ risk indicators based on World Bank,
OECD, and Ukraine's Dozorro methodology.
The system flags: single-bidder tenders (specs written for one company), repeat winners, price anomalies, bid clustering, connected companies,
and specification rigging.
Tech stack: Next.js, FastAPI, PostgreSQL, Scrapy + Playwright for scraping, Gemini embeddings for semantic search, Python ML pipeline with
150+ features.
We scrape e-nabavki.gov.mk (the official procurement portal), extract PDFs with OCR, generate embeddings, and run risk scoring. Already used
by 4,500+ companies and citizens.
The corruption detection is the interesting part - we use materialized views to pre-compute risk scores across 8 flag types, then combine them
into an overall risk rating. Happy to answer any technical questions.
