SignalFlow-AI

SignalFlow AI orchestrates procurement fraud detection through UiPath Maestro BPMN, coordinating 4 AI agents, maintaining full audit visibili, and keeping humans in control of every critical decision.

Comment

Inspiration

Procurement fraud costs enterprises over $4.7 trillion annually, yet most detection tools are black boxes — flagging anomalies without explanation, forcing operators to approve or reject blindly. We wanted to build something different: a system where AI assists but humans retain final control, with every decision fully explainable and auditable.

What it does

SignalFlow AI is a procurement fraud incident orchestration platform built on UiPath Maestro. It coordinates 4 coded AI agents through a BPMN workflow:

  • Ingestion Agent — validates incoming procurement data structure
  • Context Agent — checks vendor history from database
  • Risk Agent — calculates risk score (0-100) with Gemini/Groq AI reasoning
  • Reporting Agent — generates immutable audit trail

Every incident triggers a UiPath Maestro job. When risk score ≥ 80 (CRITICAL), the BPMN workflow pauses and surfaces full context to a human operator — who makes the final call via the Mission Control dashboard.

Key features:

  • Threat Matrix — explainable AI with per-factor risk breakdown
  • Human-in-the-Loop Override — operator approve/reject with full reasoning visible
  • Immutable Audit Log — every action timestamped and compliance-ready
  • Dual LLM Resilience — Gemini 2.5 Flash primary, Groq (LLaMA 3.1) fallback
  • Bulk Action Safety Lock — prevents accidental approval of high-risk items

How we built it

The backend is a FastAPI Python service hosting 4 coded AI agents. Each agent is a modular Python class — making them independently testable and replaceable. The agents communicate through a shared data pipeline, with results persisted to Supabase PostgreSQL.

UiPath Maestro orchestrates the entire workflow via BPMN. Every /analyze API call triggers a Maestro job through the Orchestrator API, with the process state tracked in real-time. The BPMN gateway conditions route incidents to human review or auto-approval based on risk score.

The frontend is a Next.js + Tailwind + Framer Motion dashboard designed as a cinematic Mission Control — real-time incident feed, animated Threat Matrix, terminal-style audit log, and human override controls.

AI reasoning uses Google Gemini 2.5 Flash as primary LLM, with automatic fallback to Groq when Gemini is unavailable — ensuring the system remains operational even during API outages.

Challenges we ran into

  • Dependency conflicts between google-genai, gotrue, and postgrest requiring careful version management for production deployment
  • UiPath API authentication — understanding the correct OAuth2 client credentials flow for the staging environment took significant debugging
  • BPMN condition expressions — Maestro requires specific condition expression syntax that differs from standard BPMN
  • Real-time UI updates — balancing polling frequency with API rate limits while maintaining the "live" feel of the dashboard

Accomplishments that we're proud of

  • Full end-to-end pipeline from data ingestion to UiPath Maestro job trigger in under 2 seconds
  • Explainable AI implementation that makes risk decisions transparent to non-technical operators
  • Production-grade resilience with dual LLM fallback and graceful exception handling
  • A dashboard that genuinely feels like enterprise software, not a hackathon prototype

What we learned

  • UiPath Maestro's BPMN orchestration is powerful for coordinating multi-agent workflows with human decision points
  • Explainable AI is not just a nice-to-have — it's essential for enterprise adoption of AI in high-stakes decisions
  • Dual LLM architecture significantly improves system reliability in production

What's next for SignalFlow AI

  • Real-time BPMN visualization showing live process state in the dashboard
  • PDF/invoice document parsing using UiPath Document Understanding
  • Multi-tenant support for enterprise deployment
  • Integration with ERP systems (SAP, Oracle) via UiPath API Workflows

Built With

Share this project:

Updates

Submission history