Inspiration
Traditional corporate reimbursement cycles are notoriously slow, often forcing employees to wait weeks to get their own money back. In emerging markets, this delay is compounded by volatile currencies and high banking fees. We saw an opportunity to use blockchain sub-second finality and AI agents to bridge this gap, turning a bureaucratic nightmare into an instant, transparent process.
What it does
Reimburse.ai is an AI-powered B2B platform that automates expense management and settles payments instantly on-chain.
-Instant Settlements: Employees upload receipts and receive USDC in their digital wallets immediately upon validation.
-Autonomous Auditing: An AI "Auditor" agent checks every submission against specific company policies and fraud patterns.
-Receipts Advance: A liquidity bridge that advances funds to employees even when a company’s cash flow is tight.
How we built it
We architected a dual-agent system on the blockchain:
-Agent A (The Auditor): Uses OCR and LLM logic (Gemini Live API and Gemini 3 Flash) to validate receipts against configurable rules like date windows and merchant categories. This agent is paid through x402 protocol from companies per each audit.
-Agent B (The Treasurer): Manages the automated reimbursement of USDC using secure smart contracts.
-Smart Contracts: Built with OpenZeppelin standards to handle wallet logic and payment gating. -Frontend: A submission interface for employees and an admin dashboard for companies to set spending caps.
Challenges we ran into
-AI Reliability: Mitigating "hallucinations" where the AI might misinterpret a receipt; we addressed this by designing a confidence-score threshold that flags low-certainty items for human review.
-Corporate UX: Recognizing that many companies are new to crypto, we implemented for Account Abstraction to simplify wallet management and plan "Fiat-in" services to lower the barrier to entry.
-Security: Managing automated payouts requires extreme caution, leading us to implement strict ReentrancyGuard protections and access controls.
Accomplishments that we're proud of
-Successfully designing and implementing a full-stack system that leverages blockchain speed, x402 protocol (agents payment) and live agents to outperform traditional reimbursement rails.
-Successfully generate a live, real-time and intelligent agent that talk with users meanwhile the main app processes are working.
-Developing a functional end-to-end flow that proves instant on-chain reimbursement is viable.
What we learned
We learned that the true power of blockchain in B2B isn't just the currency, but the programmability of the settlement. We also realized that while AI can handle the bulk of administrative work, a "human-in-the-loop" contingency is essential for maintaining financial compliance and trust.
What's next for Reimburse AI
Our roadmap for 2026 includes:
Q2: Expanding the policy engine to include vendor whitelists, liquidity advance and dynamic pricing.
Q3: Risk mitigation.
Q4: Integrating KYB/AML providers and undergoing full security audits for a Mainnet launch.
Scaling: Onboarding our first pilot customers to transition from prototype to a production-ready platform.
Built With
- account-abstraction
- agent-orchestration
- blockchain
- geminiapi
- next
- on-off-ramps
- python
- railway
- solidity
- supabase
- x402-protocol
Log in or sign up for Devpost to join the conversation.