Inspiration

International trade finance is still powered by manual document verification, repetitive compliance checks, and fragmented workflows. Banks and trade finance teams often spend hours reviewing commercial invoices, purchase orders, bills of lading, and other trade documents before making financing decisions. We were inspired by the opportunity to combine Agentic AI with modern financial workflows to reduce manual effort while keeping humans in control of critical decisions. we wanted to build an AI-native solution that aligns with the future of international financial services.

---

What it does

TradeFlow AI is an Agentic AI platform that automates the analysis of cross-border trade documents. Users upload trade documents such as Commercial Invoices, Purchase Orders, and Bills of Lading. Multiple AI agents collaborate to extract structured information, verify document consistency, detect potential anomalies, assess transaction risk, and generate an explainable executive report with evidence-backed recommendations. Instead of replacing financial professionals, TradeFlow AI acts as an intelligent copilot that accelerates decision-making and improves operational efficiency.

---

How we built it

We built TradeFlow AI using a modern AI-first architecture.

Frontend

• Next.js
• React
• TypeScript
• Tailwind CSS
• shadcn/ui
• Framer Motion

Backend

• Cloudflare

AI Stack

• LangGraph for multi-agent orchestration
• Retrieval-Augmented Generation (RAG)
• Structured JSON outputs

Document Processing

• PyMuPDF
• pdfplumber
• OCR support for scanned documents

Each AI agent performs a specialized responsibility, passing structured outputs to the next agent until a final risk assessment and executive report are generated.

---

Challenges we ran into

One of the biggest challenges was designing reliable collaboration between multiple AI agents instead of relying on a single LLM response. We also had to create a workflow that produces explainable recommendations rather than opaque answers. Extracting consistent information from trade documents with different layouts required careful prompt engineering and structured data extraction. Another challenge was balancing automation with human oversight, ensuring the system supports analysts instead of replacing their expertise.

---

Accomplishments that we're proud of

• Built a working multi-agent AI workflow instead of a simple chatbot.
• Created a clean enterprise-grade interface focused on usability.
• Automated cross-document verification between multiple trade documents.
• Generated transparent, evidence-backed risk assessments.
• Designed an AI workflow that demonstrates practical value for trade finance teams

---

What we learned

Building Agentic AI requires more than connecting an LLM to a document. We learned that breaking complex financial workflows into specialized AI agents significantly improves transparency, modularity, and maintainability. We also learned the importance of structured outputs, explainability, and human-in-the-loop decision-making when designing AI systems for high-trust industries like finance.

---

What's next for TradeFlow AI

Our next milestone is to transform TradeFlow AI into a complete trade finance intelligence platform. We plan to support additional trade documents such as Letters of Credit and Insurance Certificates, integrate with banking and trade finance systems, incorporate real-time shipment and foreign exchange data, enhance fraud detection with machine learning, and build collaborative review workflows for enterprise teams. In the long term, we envision TradeFlow AI becoming an AI operating system for international trade finance that helps financial institutions process transactions faster, reduce operational risk, and make smarter financing decisions with confidence.

Built With

• cloudflare
• node.js
• react
• sadcn
• tailwind