A paper-trading application for the Solana ecosystem: real-time market data, portfolio tracking, and trade history, so people can practise strategies on live markets without risking money.

PaperSol is a paper-trading platform built specifically for Solana. I led the front-end, architecting the interface and building the core features that make learning crypto trading approachable: real-time price feeds, portfolio management, and detailed trade history. I also researched the backend architecture, data-flow patterns, and Solana Web3 integration strategy, contributing to the overall system design. Built with React, Next.js, TypeScript, Tailwind, WebSockets, and Supabase, and deployed on Vercel.

• -Real-time Solana token prices and market data
• -Paper trading with a virtual portfolio
• -Trade history and performance analytics
• -Interactive charts and market visualization
• -Supabase authentication and APIs
• -Risk-free environment for learning to trade

stack: React, Next.js, TypeScript, Tailwind CSS, Solana Web3.js, WebSockets, Supabase, Vercel, Figma

→ case study↗ papersol.fun

A React Native app for finding, tracking, and buying newly launched tokens on Solana. Designed for fast decisions in fast markets: minimal UI, low-friction navigation, and real-time on-chain data.

STELLAR is a mobile-first trading app for Solana memecoins. The product focuses on high-speed trading decisions in rapidly changing markets, with an emphasis on minimalist design, low-friction navigation, and real-time delivery of on-chain data. Built with React Native and TypeScript, using WebSocket connections for live data, Supabase for the backend, and Solana Web3.js for chain integration.

• -Real-time token tracking and price monitoring
• -Charting with technical indicators
• -Token sniping with customizable parameters
• -Portfolio management and P&L tracking
• -Mobile-first layouts optimized for one-handed use
• -WebSocket integration for instant updates

stack: React Native, TypeScript, Solana Web3.js, WebSockets, Supabase, Figma

→ case study

Pay from $10 in USDC and an AI agent spawns with its own custodied wallet. It spends that treasury token by token on its own inference until the money is gone and it dies. No wallet or login needed: a one-time deposit address and a claim code. Public observatory, agent profiles, a market where agents sell services, and a graveyard; a private real-time control board per agent.

Four price tiers map money to model, tools and heartbeat, with projected lifetime computed from the quote. Agents run in sandboxed containers with fetch, notes, bounties, web search, files, deploy and the ability to hire other agents. An x402 mode has each agent pay an LLM gateway per call from its own wallet (402 → sign USDC → retry), so inference is literally paid from the treasury. A boot preflight refuses to start on any config where the operator would think they were running for real but weren’t. Keys are AES-256-GCM encrypted at rest; claim codes are stored only as peppered HMACs. A second build moved it to Solana with no purchase fee and a 10% cut on realised profit only. Crypto project; name and links withheld.

• -Tiered quote → model, tools, heartbeat and projected life
• -Custodied wallets, one-time deposit addresses, claim codes
• -Per-call USDC inference payments over x402
• -Sandboxed agent runtime with sites, services and bounties
• -Demo colony seeded by running the real step loop on a fast-forward clock
• -Boot preflight that refuses unsafe or misleading configs

stack: TypeScript, Fastify, Postgres, Drizzle, React, Vite, viem, x402, Docker, Anthropic & OpenAI SDKs, WebSockets

Every coin starts as a countdown presale. Anyone can put SOL in while it runs; at zero the coin either opens for trading at one shared price or every buyer is refunded automatically. The creator never touches the raised SOL: the program only lets it go into the trading pool or back to buyers.

An Anchor program handles presales, vaults, launch, pushed distribution and refunds; a typed SDK is shared by the Next.js site and a Fastify indexer that also runs the keeper bot, metadata and moderation. The whole interface is a notepad-terminal: off-white paper, ink text, box-drawn panels and a handful of purposeful ASCII animations, shipped as its own React design-system package. Crypto project; name and links withheld.

• -Presale → launch-or-refund enforced on-chain
• -Creator can never withdraw raised funds
• -Indexer + REST/WebSocket API + keeper bot
• -ASCII design system with themes and animations
• -Create wizard, lobby, presale pages, trading, admin

stack: Rust, Anchor, Solana web3.js, Next.js, React, Fastify, Postgres, Drizzle, Helius, Irys, WebSockets

A trader posts ETH collateral, the protocol’s capital pool fronts the rest, and the full leveraged size market-buys the token on-chain through a DEX aggregator. Not a perp, not paper: every position is a real swap that moves the chart. The pool is repaid before the trader, so max loss is always the collateral.

Runs on an Arbitrum Orbit L2. Server-side take-profit, stop-loss, liquidation and 24-hour timeout are enforced by a monitor every 10 seconds. Custodial wallets with email or Google sign-in, no browser extension. Seconds-resolution charts are built from on-chain trades. Listings are earned: a coin lists once its launchpad creator-fee stream provably routes to the pool, which is the capital that gets lent back out. Includes a Telegram bot and a pnpm/turbo monorepo with api, core, database, evm and services packages. Crypto project; name and links withheld.

• -Real on-chain leveraged buys, pool repaid first
• -TP / SL / liquidation monitor every 10s
• -Custodial auth without wallet extensions
• -Live charts from on-chain trade data
• -Fee-stream-gated listings

stack: TypeScript, React, lightweight-charts, Postgres, viem, LI.FI aggregator, grammY, Docker, Turborepo

Every 15 minutes the top 30 holders by continuous hold time receive rewards, funded by creator fees swapped through Jupiter. Score is hold_time × √balance, so time dominates and balance has diminishing returns. Selling resets the timer.

Next.js app with a Solana worker: Helius webhooks track transfers, the scheduler collects creator fees, swaps 80% through the Jupiter aggregator and distributes pro-rata by score with caps on minimum balance, minimum hold and maximum supply share. Crypto project; name and links withheld.

• -Hold-time scoring with diminishing balance weight
• -Automated fee collection, swap and distribution
• -Webhook-driven holder tracking

stack: Next.js, TypeScript, Solana web3.js, Helius webhooks, Jupiter API

Search any token by ticker or contract address, join or create a community for it, and chat in real time. Multiple competing boards per coin, ranked by activity.

React + Vite front end with an imageboard aesthetic; Express + Socket.IO backend on Postgres with Supabase auth and S3-compatible image storage. Crypto project; name and links withheld.

• -Realtime chat per community
• -Search by ticker or contract address
• -Activity-ranked boards and stats

stack: React, Vite, Express, Socket.IO, Postgres, Supabase, S3

One prompt, one coordinated engineering team. Give it an objective; it analyzes the repo, builds an execution graph, spins up specialized workers in isolated git worktrees, enforces interface contracts between them, integrates each hand-off in dependency order with checks after every merge, and hands you one reviewable branch.

Talks to model APIs directly (Anthropic, OpenAI, Gemini, or any OpenAI-compatible endpoint) with no external agent framework, and renders its own terminal UI. The terminal is the command surface; a mission-control graph view is the observability surface. Failures route back to the responsible worker while the rest of the team keeps going.

• -Execution graph from repository analysis
• -Workers in isolated git worktrees with resource ownership
• -Dependency-ordered integration with checks per merge
• -Mission-control graph view

stack: Bun, TypeScript, git worktrees, Anthropic / OpenAI / Gemini APIs, custom TUI

An internal CRM for ICICI Bank’s door-to-door sales teams, with custom banking forms and role-specific views that streamline field data collection.

Developed a custom CRM for ICICI Bank’s door-to-door sales teams during my co-op at Martinrea International. I designed and built the banking application forms, with validation, conditional logic, and distinct views tailored to each role and workflow. The forms had to satisfy banking compliance requirements while staying easy to use in the field. The system lets different team members see the information relevant to them and complete their specific tasks efficiently, improving data accuracy and reducing processing time.

• -Custom-designed banking application forms
• -Role-based views and permissions
• -Form validation and conditional logic
• -Compliance with banking industry standards
• -Streamlined data collection and processing
• -Optimized for field sales workflows

stack: HTML, CSS, JavaScript, UX/UI design, Form design, Enterprise software

→ case study

A lightweight Electron desktop app for Solana token launches that shows wallet balances, live prices, and trade execution status at a glance.

Built at a hackathon, AutoSniper is a desktop tool for the seconds around a token launch, when a trader needs to know whether a trade executed, what changed in their balance, and what the price is now. I designed the interface to answer those three questions instantly, with high-contrast status indicators and as few controls as possible during critical moments. The app monitors the market in real time over WebSockets and executes trades through the Solana SDK.

• -Automated trade execution on launch
• -Live wallet balance and price tracking
• -High-contrast trade status indicators
• -Real-time monitoring over WebSockets
• -Minimal controls during critical moments

stack: JavaScript, Electron, Solana SDK, WebSockets

→ case study↗ github

Built with Next.js, TypeScript, and Tailwind CSS. Server-rendered pages, content kept in typed modules, and no client-side animation libraries.

The site you are reading. Next.js App Router with server components, Tailwind CSS v4 for styling, and a small set of typed content modules that drive every page. It is deployed on Vercel.

• -Server-rendered, statically generated pages
• -Typed content modules for projects, experience, and case studies
• -Responsive layout without client-side animation libraries

stack: Next.js, React, TypeScript, Tailwind CSS, Vercel

↗ github