
5
1
1 year of experience
I am a BSCS graduate building strong foundations in Deep Learning with the goal of becoming an AI researcher. I focus on both theoretical understanding and practical implementation of modern AI techniques, and I am passionate about research-driven learning and innovation.

Trustless AI Agent is a demonstration of agentic commerce, showing how autonomous AI systems can transparently price tasks and settle payments in USDC on the Arc blockchain using Circle’s developer infrastructure. The project addresses a common problem in AI services: pricing trust. Users often do not know how much a task should cost, while developers struggle to implement fair, usage-based billing. This challenge becomes more important in agentic systems, where AI agents act autonomously and interact with financial infrastructure. In this system, a user submits a task through a simple frontend. The backend calls Gemini to analyze the input and classify its complexity as LOW, MEDIUM, or HIGH. Based on this classification, the AI suggests whether the task should be free or paid and proposes a USDC amount. To keep the system trustless, the AI never directly controls payments. The backend enforces deterministic guardrails using a fixed pricing table. Even if the AI output is incorrect or attempts to overcharge, the backend corrects it before any transaction occurs, ensuring predictable and auditable behavior. If payment is required, the backend executes a USDC transfer using Circle Developer-Controlled Wallets, settling the transaction on Arc Testnet. Wallet balances can be queried to verify that funds were transferred correctly, demonstrating real onchain settlement. Circle Product Feedback This project uses Arc Testnet, USDC, and Circle Developer-Controlled Wallets. Circle Wallets enable secure, programmable wallet management suitable for autonomous agents, while USDC provides stable pricing for pay-per-task models. Wallet creation, balance checks, and transfers worked well. The main challenge was limited Arc-specific end-to-end examples. More agent-focused reference implementations would improve the developer experience.
24 Jan 2026