The problem Zarfa solves

The Problem It Solves

Every year, over $200 billion in VAT refunds go unclaimed because tourists miss flights waiting in refund queues, lose receipts, or face complex paperwork. Similarly, global payrolls remain slow and expensive due to high bank fees, compliance hurdles, and multi-day transfers.

Zarfa solves both problems through a single on-chain payment infrastructure built on the Monad blockchain (EVM-compatible).

For Tourists

Zarfa lets travelers claim VAT refunds instantly through MetaMask, without standing in long airport lines or worrying about missing flights. Refunds are processed securely on-chain, and funds arrive in seconds — anytime, anywhere.

For Employers

Zarfa also enables instant, low-cost global payroll. Employers can upload a simple CSV, and the system’s AI assistant calculates, validates, and executes bulk payments across borders — ensuring transparent, compliant, and near-instant payouts.

Why It’s Better

• Eliminates paperwork and manual approvals
• Guarantees fast, secure, auditable transactions
• Uses AI + blockchain for trust, speed, and simplicity

In short, Zarfa makes money move like the internet — instant, borderless, and transparent.

Challenges we ran into

Challenges I Ran Into

While building Zarfa and the Non-Custodial Portfolio Automation Platform, I faced several technical and architectural hurdles that shaped the final design.

One of the biggest challenges was bridging off-chain AI logic with on-chain execution. My early design pushed too much logic on-chain, making smart contracts complex and costly to run. I restructured the system into a hybrid model, where the AI handles validation and computation off-chain, while only verified actions are executed on-chain — improving both performance and cost efficiency.

Another challenge was gas optimization during frequent strategy checks that ran every five minutes. The constant on-chain calls initially led to high network fees and congestion. I solved this by batching transactions and leveraging Monad’s parallel execution engine, significantly reducing gas costs and latency.

Ensuring non-custodial automation was another hurdle. Automating user strategies without taking control of funds was complex. Integrating MetaMask DeleGator smart accounts enabled delegated execution while preserving full user ownership.

Lastly, users often entered vague natural language commands like “buy when the price drops.” I built a validation layer that clarified intent, processed language context, and converted plain text into structured, executable rules — bridging human language and smart contract logic.