Yash.Arya

01

2026

Paigam

A digital invite platform: one link for the whole celebration.

A paid product I founded and built end to end. Guests get one link instead of a PDF: it opens like a real invitation on any phone, carries RSVP, events, venues and a gallery, and stays editable after it is published. Weddings first, birthdays now live, with more occasions in the pipeline.

• Self-serve studio: pick a template, type names, dates and venues, and the invitation renders live as you type. No designer in the loop.
• 17 templates across Sikh, South Indian, Muslim, Christian, Bengali, Gujarati, Royal and Minimal traditions, all driven by the same rendering engine guests open.
• Clerk auth, customer dashboard, and paid tiers at ₹99 for a single ceremony and ₹499 for up to 8 events, each with its own time and venue.
• Every invitation publishes to its own slug and opens as a rich card in WhatsApp, the distribution channel the whole product is designed around.

02

2026

PromptLock

Agent-aware distraction control for VS Code.

A VS Code extension that watches GitHub Copilot agent activity through in-memory editor events. When you're typing, your browser windows close. When the agent is working, they open. When it needs your approval, they close again. Earn the scroll, then put it down.

• Two-state machine driven by 7 detection signals: primary is onDidStartTerminalShellExecution.
• Distraction windows open at iPhone 14 Pro dimensions (393×852) with a mobile UA so Shorts and Reels scroll natively.
• Hard Lock, Punishment, and Reverse-Dopamine modes. All analytics local: zero external transmission.
• Replaces a polling approach (v1–v6) that had 17–55s detection delays. Fires instantly.

03

2026

AI Study Tutor

A subject tutor with four teaching modes and live token streaming.

A Gemini-powered tutor that adapts how it teaches: Explain, Socratic, Revision, or Exam-Focused. Tokens stream over SSE. Voice input drives a 48-bar Web Audio visualizer. Drop an image, get multimodal analysis.

• Four teaching modes: Socratic mode never gives the answer.
• Token-by-token SSE streaming with Framer Motion typewriter character animation.
• LaTeX, code highlighting, exportable notes. Zustand-scoped session memory.
• Strict TypeScript (zero any), comprehensive Jest suite, IP-based rate limiting at the edge.

04

2025

XAI-Chain

Explainable AI for blockchain transaction security.

A full-stack Web3 + AI platform that classifies blockchain transactions as malicious or benign, explains every prediction with SHAP values, and stores both on-chain via Polygon Amoy testnet. Every verdict is auditable.

• XGBoost on 9 transaction features → 100% test accuracy on 10,000 synthetic transactions.
• SHAP per-feature contributions uploaded to IPFS via Pinata, hash + risk score pinned on-chain.
• Dockerized FastAPI + Next.js 14 frontend. 6/6 Hardhat smart-contract tests passing.
• Three-page frontend (landing, analyze, dashboard) with RainbowKit + Wagmi v2 wallet auth.

05

2026

Bezel

A component system with an accessibility bar it has to clear.

A design system published as readable TypeScript source, with a CLI that copies components straight into consumer projects. Tokens for colour, type, spacing and motion, a documented states matrix, and a contrast gate that fails the build on WCAG AA violations.

• bezel add <component>: source in your repo, not a dependency you fight.
• Visible keyboard focus on every interactive component, enforced not assumed.
• Registry generated from source, so the gallery can never drift from what ships.

06

2025

HAM10000 Skin Lesion Classifier

Deep learning on 10,015 dermoscopy images, 7 lesion classes.

PyTorch pipeline classifying skin lesions with EfficientNetV2-RW-S and ConvNeXt-Tiny ensembles. Focal loss handles a brutal 6,705-vs-115 class imbalance. Grad-CAM heatmaps make every prediction inspectable.

• EfficientNetV2-RW-S → 94.21% validation accuracy on a single RTX 3060 in ~1.5h.
• Focal Loss (γ=2.0) for class imbalance · Mixed Precision FP16 for 2× speedup.
• AdamW + CosineAnnealing LR + best-only checkpointing. Stratified 85/15 split.

07

2024

ShoeStore

Headless e-commerce with Stripe, Cloudinary, and a Strapi backend.

Next.js storefront on a Strapi CMS. Stripe for payments, Cloudinary for media, Redux for cart state. Mobile-first responsive with a measured 25% engagement lift on mobile.

• Stripe checkout + Cloudinary-optimized media + Strapi-managed catalog.
• Redux cart state with persistence across sessions.
• Mobile-first responsive: 25% engagement boost vs. desktop-first prototype.

08

2024

Crypto Portfolio DApp

Multi-chain portfolio with cross-border payments.

MetaMask + RainbowKit wallet auth, ERC20 + NFT tracking with on-chain queries, live pricing via CoinMarketCap, OpenAI-generated portfolio insights. Three payment rails: Ethereum, card, bank.

• Cross-border payments authenticated by Email + Mobile OTP.
• Live ETH price feeds + multi-step checkout.
• ERC20 and NFT inventory queries against multiple chains.