Risk scoring on four axes, detection of collusion between bidders (shared beneficial owners, rotating winners, overlapping IP addresses and document metadata, complementary pricing), an immutable hash-chained log of every system decision. Licensed separately — to the Antimonopoly Service, the Chamber of Accounts and SOE auditors.
A procuring entity should not see the tools that audit its own procedures for collusion signals. A bidder should not see audit signals about their own bids. That is why AZTender Audit is a separate product with its own access, its own database and an independent delivery.
Six AI modules, each licensable separately or as a bundle. All built and validated against live etender.gov.az data (105,000+ AZ tenders) and ProZorro (100,000+ UA tenders for cross-country benchmark).
Risk level, collusion probability, fairness, competition level — each scored 0–100 with an explainable breakdown. Z-score for price anomalies, dumping detection, domain rules tuned for Azerbaijani procurement.
Beneficial-owner overlap by VÖEN, shared phone numbers and addresses, overlapping IP addresses and document-creator metadata, rotating winners, complementary pricing, suspicious bid withdrawal. Each pair carries a confidence score.
Every decision and every user action is written to a log that is physically immutable — a Postgres trigger blocks updates and deletes. Aligns with EU AI Act art. 12 (record-keeping). Chain integrity is verifiable in a single query.
EU, UN, OFAC, UK lists, AR debarment registry, specialized FinCEN lists, Interpol notices, sectoral sanctions. Match by name, alias and VÖEN with deduplication. Daily refresh.
Every flagged decision requires confirmation by a human auditor. The confirmation is written to the log with user and timestamp. Ready-made legal infrastructure for regulator use — nothing to build on top.
Every score keeps the model name, model version, feature-set version and input hash. A year later you can re-run the same data through the same weights and get an identical result — exactly what oversight bodies demand.
A regulator's installation lives at audit.aztender.ai (or your own subdomain, or inside your infrastructure) with its own database. Your data is physically separated from the public side of the platform — no overlap.
If the law requires data not to leave the country — we deploy on your server. Containerized, single-binary delivery, ~30 minutes to roll out.
Webhooks for every significant event — high risk score, collusion match, sanctions hit. Your data sources can be plugged in: internal registries, court judgments, e-justice records. REST API with OpenAPI documentation.
Articles 12 (record-keeping), 13 (transparency), 14 (human oversight) and 15 (accuracy & reproducibility) — all four high-risk AI requirements are baked into the architecture, not bolted on.
Demos are conducted under a non-disclosure agreement for authorized staff of the regulator, SOE auditors, and forensic teams of partner audit firms (Big 4). Each session is 60 minutes: 30 min architecture review, 30 min live demo on anonymised data.
Response within 1 business day. NDA signed within 24 hours of the first reply.