Which Plushie Email Verification Design Do You Prefer?
I explored a couple of directions for Plushie’s email verification experience, keeping the core idea the same while experimenting with different layouts, visual hierarchy, and ways to make the verification code feel more playful.
Version 01 - Visual-first, with a strong hero illustration
Version 02 - More structured, clean, and email-focused
Both follow the same soft, plushie-inspired visual language.
Version 02, mostly because showing it inside the actual email client chrome builds more trust that this is really landing in the inbox. Version 01's illustration is charming but it's competing with the code for attention.
Project 1 — Email Agent
Title: AI Email Agent — Automatic inbox classification and reply
Description:
Intelligent agent that reads, classifies and auto-replies to emails using GPT-4o. Connects to Gmail via API, detects intent with 97%+ accuracy, generates context-aware drafts in the user's language, and requests human approval before sending. Built with Python, OpenAI API and Google Workspace APIs.
Multi-step automation system that chains AI actions across tools. Fetches emails, classifies urgency with GPT, checks calendar, writes summaries to Notion and triggers alerts — all in under 2s. Supports Gmail, Google Calendar, Drive and custom REST APIs.
Project 3 — Voice Assistant (JARVIS)
Title: JARVIS — Voice-controlled AI assistant with full system access
Description:
Personal AI assistant controlled entirely by voice. Uses Whisper for speech recognition, GPT-4o-mini as the reasoning core, and 18 custom tools to open apps, control Spotify, read emails, browse the web and manage files — all hands-free. Built with Python, persistent memory via Obsidian, and a real-time 3D HUD interface.
Project 4 — Data Pipeline
Title: DataFlow AI — Intelligent data pipeline with AI enrichment
Description:
Automated data pipeline that ingests records from multiple sources (Gmail, Drive, REST APIs), cleans and deduplicates them, then enriches each record with AI: entity extraction, sentiment analysis and intent classification. Processes 2,800+ records per minute with 99.2% accuracy. Built with Python and OpenAI/Claude APIs.