Ringfence

• 

  Built with PayPal Payouts and Agent Toolkit, plus AG Studio, Bryntum Scheduler and APIMatic. Code moves the money; AI only explains.

• 

  One front door for every caller. Only packages/paypal can move money: outbox, idempotent send, forward-only states, 5 crash tests.

• 

  The console: 387 paid on time, 23 paused as likely fraud, 2 as likely real. Group 12 shows its 3 shared devices and 12-minute burst.

• 

  Learned rules: the fire drill and the rule written from Group 12. Insights, built on AG Studio, shows money kept versus leaked.

• 

  Overview: where every payee in a payout run went, with the cases that need a decision listed by deadline. One click: fraud or not fraud.

• 

  Decisions: each held group with a graph of what links its accounts and checked findings. Choose not fraud to pay it, or fraud to block it.

• 

  Learned rules: confirming fraud writes a rule, which Ringfence attacks with 200 disguised copies first. Nothing runs until you turn it on.

• 

  Insights: an AG Studio dashboard over every decision. See money kept versus leaked by run, or ask a question and get a chart.

Inspiration

Our story is rooted in the collective, firsthand experiences of our team across multiple sectors:

• Angel: A Product Manager who has overseen a diverse range of applications in different markets and operational workflows.
• Austin: A Developer and AI Researcher with deep experience building payment products for a global enterprise.
• Chelsea: A Consultant-turned-Developer who has tackled problem spaces across various industries.

Across our backgrounds, we repeatedly saw the same systemic failure: traditional fraud handling is slow, unsustainable, and not scalable. In many systems, by the time a fraudulent transaction or anomalous behavior is flagged and processed, the damage is already done.

We built Ringfence on the conviction that fraud detection and response must be immediate. However, automation shouldn't mean losing control—effective risk mitigation requires keeping a human in the loop at all critical decision points.

What it does

Ringfence provides an immediate, intelligent defense layer against fraudulent activity across application ecosystems.

Key capability highlights include:

• Immediate Detection & Action: Evaluates incoming events in real time and automatically triggers configurable defense protocols instantly.
• Human-in-the-Loop Workflow: Enables risk operators and analysts to review, override, fine-tune, and approve actions without bottlenecking real-time protection.
• Centralized Console: Offers an intuitive dashboard for monitoring security events, analyzing risk triggers, and managing incident responses dynamically.

How we built it

We designed Ringfence with modularity and speed in mind, ensuring seamless integration between real-time data processing and operator intervention:

• Frontend & Operator Console: Built as a responsive dashboard allowing real-time visualization of risk signals and manual intervention points. Hosted live via Render at ringfence-console.onrender.com.
• Core Logic & Engine: Developed using scalable backend architectures to process telemetry and event streams asynchronously, scoring threats in real time.
• AI & Rules Engine: Leveraged machine learning models alongside configurable deterministic rules to balance fast automated actions with low false-positive rates.

Challenges we ran into

• Balancing Speed vs. Human Control: Achieving sub-second automated responses while maintaining intuitive mechanisms for human intervention required careful architecture around state management and event propagation.
• Unifying Industry Paradigms: Combining insights from payment systems, general app management, and enterprise consulting into a single, cohesive schema that works seamlessly across diverse domains.
• Streamlined Deployment: Configuring multi-tier components to run efficiently on cloud infrastructure while ensuring real-time capabilities remained performant.

Accomplishments that we're proud of

• Successfully translating a complex cross-industry problem into an operational working prototype.
• Creating a unified platform where AI-driven automated defense seamlessly coexists with human oversight.
• Deploying a functional end-to-end console and open-sourcing our core platform work on GitHub.

What we learned

• Real-time is non-negotiable: Near-real-time is often too late in high-risk environments like modern payments and online services.
• Humans remain essential: Fully automated black-box solutions fail due to lack of trust and edge-case rigidity; empower human analysts rather than replacing them.
• Cross-disciplinary collaboration drives better design: Combining PM, AI research, enterprise payments, and consulting perspectives enabled us to build a more holistic solution than any single discipline could produce alone.

What's next for Ringfence

• Enhanced Rule Engine: Expanding support for custom policy builders and complex event processing (CEP).
• Advanced AI Analytics: Integrating deeper predictive threat modeling and dynamic risk scoring based on historical behavioral patterns.
• Third-Party Integrations: Developing plug-and-play SDKs and webhooks for popular web frameworks, payment gateways, and identity providers.

Built With

• aggrid
• agstudio
• apimatic
• bryntumscheduler
• fastify
• gemini
• graphology
• next.js
• payouts
• paypal
• postgresql
• render
• rest
• typescript