All builds

Every project I've shipped — 8 builds across ZK settlement, IoT ledgers, fraud scoring, and on-chain tooling.

Every build

Money only settles once the work is provably done.

A ZK-gated USDC escrow protocol on Circle's Arc testnet. Work is verified with a Groth16 proof before funds settle — no trusted third party checking the output.

4 Solidity contracts + a Circom circuit with Poseidon hashing, proved entirely in-browser via snarkjs.

SolidityCircom / ZK-SNARKssnarkjsReact · wagmi

Live app Source

IoT devices that pay automatically when checkpoints are met.

A Hyperledger Fabric network wired to ESP32 IoT hardware — devices report status on-chain via a webhook pipeline, and payments release automatically once conditions are verified, no manual settlement step.

ESP32 devices push signed status updates through a webhook layer into a permissioned Hyperledger Fabric ledger deployed on Azure, with a chaincode-triggered payment release once each checkpoint is confirmed on-chain.

Hyperledger FabricESP32 / IoTAzureWebhooks

Live app

Deploy a smart contract to Base in one tap.

Deploy a smart contract to Base in a single tap, then track its wallet health from the same dashboard — built to strip friction out of shipping on-chain.

50+ real users deploy live contracts on Base — pick a template, sign once, and an automated pipeline verifies the bytecode on-chain, indexes the contract into a wallet health dashboard, and turns it into a shareable on-chain resume entry.

Base (L2)Solidity

Live app Source

Get pinged the moment coordinated wallets move money.

A cross-market smart-money coordination detector — traces wallet clusters moving in sync across Hyperliquid perps, then pushes real-time alerts to Telegram.

BFS traversal across wallet clusters flags synchronized moves across Hyperliquid perps, cross-referenced against Nansen smart-money labels, with alerts pushed to Telegram in real time.

Hyperliquid APINansen labelsTelegram bot

Live app Source

Blocks UPI fraud before the money settles.

A pre-settlement UPI fraud risk scorer. An Investigator agent and an Analyst agent run every transaction through five weighted rules before it settles, not after.

A Mastra-orchestrated two-agent pipeline scores every incoming UPI transaction before settlement — an Investigator agent flags velocity and spike anomalies against transaction history, then an Analyst agent (Groq, Llama-3.3-70B) reasons over those scores to return a real-time fraud verdict.

Mastra multi-agentTypeScriptGroq / Llama 3.3

Source

A single hub for discovering everything live on Monad testnet.

A dashboard that tracks dApps deployed across Monad testnet — organized by DeFi, NFTs, and Games, with a dedicated Archive section for projects that have gone dark. Built with a teammate (Manisai); I owned the interactive discovery layer.

Built the 3D discovery visualizer for browsing apps by interest, the tipping tool, list view + favorites system, and sound design — the entire community module was mine end to end.

React / TypeScriptVite3D visualizer

Live app Source

Every frame flip is a real on-chain transaction, not just a UI state change.

An interactive 162-frame viewer where each frame advance triggers a gasless on-chain transaction — user deposits MON, then batch-views or steps through frames one at a time, with withdrawal support built in. Wired into Monad Games ID for cross-app identity and leaderboards.

Privy handles the embedded wallet and account abstraction layer (sponsoring gas via ERC-4337), integrated with Monad Games ID, while a Solidity contract deducts 0.001 MON per frame viewed on Monad testnet.

Next.jsPrivy AASolidity + OpenZeppelin

Live app Source

Feasibility reports grounded in live market data, not assumptions.

A Google ADK multi-tool agent takes a user's startup idea, target location, and funding range, then fetches real-time cost and existing-business data for that location. Instead of a single prompt-response, this is a multi-turn agent that gathers structured inputs, pulls live market data, checks back with the user at key decision points (human-in-the-loop) rather than assuming, and compiles everything into a final feasibility report.

Multi-tool orchestration in Google ADK: gathers structured inputs, fetches real-time cost and existing-business data, holds for human-in-the-loop confirmation before finalizing assumptions, and outputs a setup-cost estimate against the user's stated requirements.

Google ADKPythonHuman-in-the-loop

Source