
Argus
An AI supervision layer that sits on top of existing ERPs, catching financial and vendor-registration anomalies traditional systems miss.
Video demo
About the project
Companies running traditional ERPs share a silent problem: the system records everything but supervises nothing. Payments go out without reconciliation, vendors with irregular tax IDs keep getting paid, invoices expire without approval. These issues only surface at month-end close — too late to prevent the loss.
Argus is an intelligent supervision layer that sits on top of the ERP. It doesn't replace the existing system — it makes it smarter.
The platform runs on two specialized AI agents that operate autonomously and continuously. The Finance Agent monitors ledger entries, bank statements, and payables/receivables, flagging inconsistencies and anomalies in real time. The Registry Agent automatically queries Brazil's federal tax authority (Receita Federal), verifying the regulatory standing of vendor and customer tax IDs and catching registration mismatches.
When an inconsistency is detected, the agents generate exceptions ranked by severity and route them to the manager through the Control Tower — an executive dashboard with full analysis, supporting evidence, an audit trail, and a recommended action. The manager stays fully in control: resolve it directly, forward it to the responsible party via WhatsApp, email, or Teams, or escalate it for review.
The system is built on a multi-agent, human-in-the-loop architecture, directly aligned with the hackathon's theme. Agent intelligence is powered by Anthropic's Claude API, with live integration to Brazil's federal tax API via BrasilAPI.
The business model is B2B, using ERP implementation consultancies as the primary distribution channel — turning a one-time implementation engagement into a recurring intelligent-monitoring retainer.
Team Argus Lab
Arthur Souza
Rian Gabriel de Oliveira
Pedro Lima
Maitê Nascimento de Paula Silva
