• 

  Manual Verification → Autonomous Triage （Before / After）

• 

  Multi‑AI Consensus Architecture

Focus Shield AI: Autonomous Multilingual Invoice Triage

💡 Inspiration

Our project stems from a raw, real-world pain point experienced by global business professionals: the nightmare of manual invoice processing. Managing an expanding volume of invoice PDFs every month across diverse languages (Japanese, English, German, Spanish, Thai) and chaotic formats is deeply draining. This operation is rarely a staff member's core duty—it is squeezed into the gaps of their daily schedule, stripping away their focus and cognitive energy. We realized that manual full-volume verification is fundamentally unsustainable. We needed an "AI-Led Shield" to silence the operational noise and empower human genius.

🛠️ What it does

Focus Shield AI is an Autonomous AI Triage System that shifts the workflow from human validation to AI-driven exception handling. Instead of forcing humans to check every single invoice, the system cross-examines data extracted by multiple AI models.

It instantly registers flaw-free invoices directly into the database (Straight-Through Processing). If and only if a mismatch or anomaly is detected, the system intelligently isolates the case and flags it for human review. Humans interact strictly with the "exceptions," fully restoring their focus and workplace morale.

🏗️ How we built it

We built an End-to-End autonomous pipeline by fully leveraging the UiPath Platform Ecosystem:

1. Trigger & Queueing: UiPath Integration Service/RPA monitors incoming emails. Upon receipt, a dispatcher downloads the invoice PDFs and registers metadata (Subject, Body, Path, Sender, Mail ID) into Orchestrator Queues.
2. Data Fabric Integration: Orchestrator Queue triggers an RPA bot to immediately log the incoming queue data into UiPath Data Fabric records.
3. Multi-AI Consensus & Triage (Core Innovation): A specialized AI Agent uses Intelligent Document Processing (IDP) to read the uploaded PDFs, identify languages/formats, and update the Data Fabric records. Crucially, the agent acts as an autonomous triage manager, cross-referencing values to detect errors.
4. Human-in-the-Loop: If an anomaly is found, the agent automatically creates a task in UiPath Action Center and writes the Action Center URL back to the Data Fabric record. The user reviews, fixes, or approves the exception via a unified interface.

5. Visibility & Governance: Users can monitor the end-to-end processing status at any time via a custom dashboard built with UiPath Apps.

🏆 Challenges we overcam & Accomplishments

The biggest challenge was moving away from the traditional, rigid "AI extracts, human checks everything" model. By shifting to an "Agentic AI Triage" model, we engineered a system where distinct AI insights validate one another, minimizing human touches. We successfully proved that even the most chaotic, multilingual, and multi-format global financial workflows can be neutralized and automated smoothly under the UiPath umbrella.

🚀 What's next for Focus Shield AI

Focus Shield AI is not just a tool to cut down time and effort; it is a true foundation for innovation. By shrinking a monthly 2-to-3-day grind down to just a few hours, we successfully reclaim valuable human mindshare. Looking ahead, this unified and clean data foundation opens up inspiring future possibilities. For instance, it can easily serve as a springboard for advanced creative extensions, such as integrating LLM capabilities for predictive cash-flow analytics. Our ultimate vision is to empower teams worldwide to completely break free from routine, restore their focus, and boldly venture into high-value business frontiers.

Built With

• claude-api
• fastapi
• microsoft-teams-api
• openai-api
• postgresql
• python
• rag
• react
• supabase
• tailwind-css
• typescript
• uipath
• vector-db
• vercel
• whisper-api