case study · arce tenders
Arce, in a diagram you can move around.
Four views of the system: the full circuit from the portal to the dashboard, how a batch of historical data comes in and gets normalized, how an award becomes a statistic and how the AI writes the analysis without making up figures. Every box can be dragged; every view has a step-by-step walkthrough.
From the public procurement portal to management: the history comes in through an API, gets normalized in PostgreSQL, the statistics are precomputed and an AI reads them to write the analysis. All in one Docker service; the AI runs on Ollama or Anthropic.
Two-minute script
- Arce answers a question from management: what is worth selling to the State, at what price and against whom. To do that it uses the full history of Uruguay’s public purchases, not just open tenders.
- A separate scraper, still pending, will walk the portal and send tenders in batches to an API. The system is already ready to receive them.
- The API validates each tender, stores the raw payload and normalizes it: suppliers by tax ID, articles by code, currencies and units in a single canon.
- On top of that, PostgreSQL precomputes the statistics: what to sell, at what price, who wins and who buys. An AI reads those figures and writes the report, without computing anything on its own.
- The previous stage still runs in the same service: the monitor that filters open tenders in one field and leaves them in a review inbox.
Click a box to see what it is and what it connects to. Drag to arrange; “Rearrange” restores the original layout.