Mind the Gap: Real Time Compliance

Inspiration

Compliance is the most important concern for banks, financial institutions, and any licensed entity. One failure can mean millions in fines, loss of reputation, or even loss of license. Yet compliance systems today are ingrained with legacy platforms, making it a risk to adopt new technology. We wanted to show how blockchain can be plugged into what already exists, not replace it, to provide tamper-proof, regulator-verifiable evidence of compliant operations.

What it does

Our solution creates a real-time compliance layer that anchors existing AML, KYC, fraud detection, and suspicious activity reporting (SARs) systems to blockchain. It demonstrates how banks can:

Monitor transactions and detect suspicious activity in real time
Generate SARs instantly, anchored immutably to a Cosmos chain and provide these SARs in real time to a regulator
Export daily compliance reports for regulators aggregating transactions from the chain
Bridge transactions happening on-chain, off-chain, and across both

How we built it

First we discussed how we can bridge the gap between on-chain and off-chain data, relevant to prove to a regulator a bank is compliant
One of the common barriers to move any financial use case into production is connecting off-chain identities and information relevant for compliance, or audit trails to the on-chain usage.
We researched the reporting obligations related to KYC/KYB, fraud, and AML with AI and decided to incorporate the requirements for obligations in these areas into the demo.
We laid out a user flow for the demo and rough wire frames and used figma make to generate the front end designs.
We used various AI tools: claude code, and cursor to get the demo up and running.

Challenges we ran into

Getting the generated client code based off of the protobuf schema's to work with the AI generated frontend code was a big challenge which took several hours to make it compile
One of our team members fell ill during the hackathon, we had to readjust to this reduced capacity
- Leveraged multi-agent flows to run multi-attempt workflows overnight to compensate.
Syncing work between multiple vibe codes can have challenges, for example in re-creating the same data structure multiple times, we had to re-factor this multiple times
The time constraint made things hard
AI generated video has its limitations, and usage limits even on paid versions mean you only have a few attempts to get it right

Accomplishments that we're proud of

We went through a thorough product discovery journey from ideation and research through to delivery
The contributors who are not engineers in their day job tapped into the vibe coding of the front end
We have delivered a project which is not just code, but has design, narrative and a video within it

What we learned

Fraud detection and suspicious activity reporting require cross-linking identity data with transactions. We realized the demo is most compelling when we show this linkage explicitly (KYC → flagged transaction → SAR → blockchain tx hash).
AI sped up design and coding, but we learned that compliance logic (rulesets, thresholds, reporting formats) can’t just be AI-generated — it required domain research and careful tailoring.
A dry problem becomes engaging when framed as a story: our 90s-style video helped us see how important narrative is to get people to care about compliance.

What's next for Real Time Compliance

Expand beyond B2B payments to cover broader banking workflows: trade finance, lending, escrow-based payments, and any other product offering a bank could imagine
Deeper integration with open-source AML engines and live fraud detection APIs.
Pilot with real financial institutions to prove the model in production.
Build towards the broader vision: migrating more and more banking operations onto a blockchain backbone, with compliance built-in by design.

AI Tools Used

Ideation and idea refinement, Research (ChatGPT, Claude)
Frontend/Off-chain Code (Cursor, Claude Code, Codex, Copilot)
- Multi-agent workflows to coordinate large bodies of work overnight (Cursor Background Agents, Claude Code, Copilot on GitHub)
Voice-to-text for more efficient prompting (Monologue)
Design and frontend (Builder.io)
AI Video (OpenAI Sora, Gemini)
Design and Frontend (Figma Make)
Imagery (Midjourney)