Procurement Technology / AI
TenderScanner
AI procurement intelligence platform
My own SaaS product: scrapers ingest public tenders across Serbia and Bosnia and Herzegovina, and an AI pipeline summarizes, embeds and scores each one against company profiles to produce go/no-go recommendations.
Contribution
- Everything - product, frontend, backend, data platform and AI pipeline, built solo
- Scrapers ingesting newly published tenders from the Serbian and Bosnian procurement portals into PostgreSQL
- AI pipeline - Gemini summaries, embeddings and a weighted fit score per tender and company profile
Companies bidding on public tenders in Serbia and Bosnia and Herzegovina track them the hard way: fragmented procurement portals, manual checking, and a real cost to missing the one tender that fits. TenderScanner is my answer to that, a product I design, build and run solo.
The pipeline works end to end in production. Scrapers pull newly published notices from both national portals into PostgreSQL once a day. Each new tender then passes through a Gemini Flash step that writes a Serbian-language summary and suggested bidder questions, gets embedded, and is scored against every registered company profile with a weighted fit score across industry, CPV codes, keywords, regions and value range. The dashboard turns that into a go/no-go recommendation with the supporting detail, plus favorites with change tracking.
The stack is a React and TypeScript frontend on a Node.js backend with PostgreSQL and a job queue underneath; the backend is currently being ported to .NET. It is live, with the first interested clients and a feature roadmap ahead, and it doubles as the project where I own every decision, from scraper resilience to what an AI recommendation owes its reader.
Outcomes
- Deployed and operational - scrapers and AI analysis run on every newly published tender
- First client conversations underway while development continues