VisionLedger

VisionLedger automates supply chain crisis remediation in 180s: combining a cognitive multi-agent AI swarm, secure SAP re-routing, and audit-ready ledgers orchestrated by UiPath Maestro.

Comment

Inspiration

In an era of unprecedented geopolitical friction, climate-induced natural disasters, and black-swan logistical disruptions (such as shipping chokepoints in the Red Sea or manufacturing lockdowns in Taiwan), multinational corporations face a critical vulnerability. When a supplier halts production, global supply chain teams often take weeks to react due to fragmented silos, manual emails, and outdated procurement reviews. Every day of delay costs Fortune 500 enterprises millions in operational downtime.

We were inspired to build VisionLedger: an autonomous, highly secure "Force Majeure" Case Management platform. We wanted to prove that modern enterprises do not have to choose between the cognitive intelligence of Generative AI and the deterministic, audit-safe rigor of robotic automation. VisionLedger bridges this gap, resolving complex supply chain crises in under 180 seconds under absolute human governance.

What it does

VisionLedger is a Next-Generation Supply Chain Crisis Command Center designed to automate risk triage, alternative sourcing, compliance verification, and ERP execution:

  1. Ingestion & Social Anomaly Radar: Our Sentinel AI continuously monitors global feeds. Once a disruption signal is ingested, it calculates geospatial coordinates and initiates the case workflow.
  2. Cognitive Multi-Agent Swarm (AI Brain): A collaborative swarm of specialized agents (Triage, Financial, Logistics, Temporal, and Compliance) powered by Google Gemini 2.5 Flash instantly goes to work. They analyze supply chain exposure, evaluate transit delays, search for alternative safe suppliers, and reject close-proximity candidates to avoid cascading Tier-2 dependencies.
  3. Interactive 3D War Room & Explanability (XAI): Executives visualize the crisis on a WebGL 3D globe showing dynamic red/green logistical arcs. The Decision Intelligence Sheet translates the AI's complex Monte Carlo risk probability and carbon impact into clear financial indicators (Inaction Cost vs. Remediation Cost).
  4. Human-in-the-Loop Governance: No financial transaction is committed blindly. The Sourcing Officer reviews the XAI report and clicks "Authorize Sourcing Route."
  5. UiPath Maestro RPA Execution: Upon human authorization, UiPath Maestro acts as the central state engine. It spawns a secure process robot that connects to the enterprise ERP (SAP S/4HANA), modifies the sourcing allocation, and creates the official Purchase Order (PO).
  6. Cryptographic Ledger & Auditability: The system compiles an emergency resolution PDF report signed with an immutable SHA-256 session hash and a secure QR code containing compliance metadata, archiving it automatically for regulatory standards.

How we built it

We designed VisionLedger using a strictly decoupled, asynchronous monorepo architecture:

  • Frontend (React & WebGL): Built with React and Vanilla CSS for a premium, high-fidelity dark UI. We leveraged react-globe.gl to render the interactive 3D WebGL globe, animating geographic arc vectors in real-time. We used recharts to build the predictive "What-If" cost chart and the ESG compliance gauge.
  • Backend (FastAPI & Gemini): Engine coded in Python. We integrated the Google Gemini 2.5 Flash model with robust Prompt Injection Guards to classify text anomalies and generate supplier candidate scenarios. Server-Sent Events (SSE) stream the AI Swarm's live thoughts to the client. Disk I/O operations are offloaded to background threads (asyncio.to_thread) to maintain non-blocking asynchronous execution.
  • Orchestration & RPA: We used UiPath Maestro as our supreme case orchestrator to manage state transitions and coordinate webhooks. The backend acts as a secure proxy API to communicate with UiPath webhooks without exposing tokens to the client.
  • Audit & PDF Engine: Built headless PDF generation utilizing fpdf2 and qrcode to dynamically output certified compliance ledger reports on the server.

Challenges we ran into

Integrating the highly creative, non-deterministic reasoning of LLMs with the strict, transactional rules of enterprise systems like SAP was our greatest challenge. If an AI hallucinates a non-existent supplier or proposes a sanctioned entity, it could lead to legal catastrophe.

We solved this by establishing a hybrid compliance model:

  • Hard deterministic gates (like official OFAC sanctions databases and deterministic MD5-based vendor verification algorithms).
  • Semantic AI agents for unstructured ESG assessment.
  • The ultimate safeguard: a strict Human-in-the-Loop barrier coupled with UiPath Maestro's state-machine tracking.

Another challenge was managing high-frequency WebGL rendering on the client side alongside the live EventSource SSE stream without dropping frames. We optimized this by decoupling React state updates from the WebGL globe's continuous auto-rotation loop.

Accomplishments that we're proud of

  • End-to-End Monorepo Synergy: VisionLedger is not a mock or a slide deck. The React frontend, FastAPI backend, Gemini agent swarm, and UiPath Maestro trigger webhooks communicate and execute smoothly.
  • Double Terminal Observability: The dual terminal interface—showing the AI Swarm's cognitive debate stream on the left and the UiPath robot's SAP ERP execution log on the right—provides breathtaking clarity on how AI and RPA work together.
  • The Cryptographic PDF Report: Generating a dynamic, signed PDF containing a secure QR code that immediately displays JSON audit data upon physical mobile scanning adds an extraordinary level of credibility.
  • Premium UX/UI: The glowing, cyber-inspired dark mode UI makes the crisis room feel expensive, premium, and executive-ready.

What we learned

  • AI and RPA are better together: AI is the "brain" that thinks, arbitrates, and adapts to unstructured real-world crises. RPA is the "arms" that execute transactionally with absolute precision, speed, and security in legacy systems.
  • The Power of Case Management: Designing the workflow using UiPath Maestro's case principles taught us how to construct resilient architectures where the state is preserved even if a network socket fails.
  • Explainable AI (XAI) is mandatory: Corporate leaders will never trust a button that says "Trust the AI." They need clear metrics, risk indicators, and trace logs to back up the recommendation.

What's next for VisionLedger

  • Live Intelligence Feeds: Integrating actual API streams for global transit risks (e.g., live port congestions, meteorological storm projections, and conflict zone maps).
  • Multi-Tier Approval Delegation: Implementing automated delegation of authority (DoA) thresholds, where minor reallocations under $50k are executed automatically by Maestro, while larger allocations trigger multi-step executive approvals.
  • Direct SAP S/4HANA Integration: Upgrading the simulated ERP robot to write directly into an enterprise SAP sandboxed instance using native UiPath activities for a production-ready pilot deployment.

Built With

  • fastapi
  • fpdf2
  • git
  • google-gemini-2.5-flash
  • netlify
  • python
  • qr-code
  • react
  • recharts
  • sap-erp
  • server-sent-events-(sse)
  • sha-256-cryptography
  • uipath-maestro
  • uipath-orchestrator
  • vanilla-css
  • vite
  • webgl
Share this project:

Updates