Licitiraj.ba
Public-sale discovery built from trustworthy ingestion and measurable extraction quality.
Licitiraj.ba is being built from the data layer outward. The current engineering phase validates official public URLs, deterministically extracts a minimal normalized public-sale model, blocks raw or personal content, deduplicates canonical sources and measures extraction quality against a sanitized golden set before a public discovery interface is introduced.
Product type
Data platform · public-sale discovery
Status
In development · data foundation
Astraealibra scope
Product design · ingestion architecture · data quality · benchmark engineering
Product identity
Licitiraj.ba
Product in practice
The product proof currently lives in the data system.
Licitiraj.ba does not yet have a public interface. Instead of inventing an application mockup, this portfolio shows the real development flow the product currently depends on — from official sources to verified, measurable data prepared for a future public experience.
Domain
Licitiraj.ba
Access
Data foundation
Current product flow
Functional scope
What the product is built to do.
Validation of official public-sale source URLs before ingestion
Deterministic extraction into a minimal normalized sale model
Privacy boundaries that reject raw and personal content from persistence
Canonical URL deduplication across repeated or overlapping notices
Offline extraction-quality benchmarking against a sanitized golden set
Database-level visibility rules for active and non-expired sales
Manual provider proof path kept outside normal tests and CI
System architecture
How the product is structured beyond the interface.
Source boundary
01Only approved official public sources enter the ingestion path, with explicit URL validation before extraction work begins.
Extraction & privacy
02Deterministic validators and extractors produce the minimum public-sale model, reject prohibited raw/personal content and deduplicate canonical URLs.
Quality benchmark
03A sanitized 50-notice, 12-source golden set drives an offline extraction benchmark so quality changes can be measured before they reach production.
Persistence & visibility
04TypeScript services use Drizzle ORM with PostgreSQL/Neon; the active-sales database view applies active-status and expiry rules at query time.
Engineering highlights
The work that makes the system durable, testable and operable.
Privacy validation is a storage boundary, not a cleanup step after persistence
Deterministic extraction is benchmarked against frozen expected results instead of judged by anecdotal examples
Normal tests, build and benchmark are credential-free and make zero Firecrawl and zero Neon network calls
Manual provider proof is isolated from CI and never writes raw provider content to the production database
Active-sale visibility is enforced in the database query model rather than left to frontend filtering
Technology
Production stack and supporting infrastructure.
Current direction
Where the product is now.
The current phase deliberately excludes the public frontend. Work is focused on trustworthy ingestion, privacy-safe normalization and measurable extraction quality before discovery UX is added.