Cubby

The Situation

Cubby had a compelling vision for an AI-powered email intelligence platform, but they were racing against the clock. Investor conversations were heating up, and they needed more than slides—they needed a working product to demonstrate traction.

Their previous development attempts had stalled. Estimates ranged from 3-6 months for an MVP, which would have burned through their pre-seed runway before they could even validate the core hypothesis.

The Constraints

• Timeline: Investor demo scheduled in under 3 weeks
• Budget: Pre-seed resources, needed to maximize runway
• Technical: AI parsing needed to be accurate enough for real user testing
• Scope: Full-stack application with authentication, dashboard, and AI processing

The Approach

We took a "validation-first" approach—focusing on the core value proposition while building on a foundation that could scale.

Week 1: Foundation & AI Pipeline

• Set up Next.js project with TypeScript for type safety
• Integrated Claude API for email parsing with structured output
• Built the core parsing logic with proper error handling and fallbacks

Week 2: Full-Stack Integration

• PostgreSQL database schema optimized for email data
• User authentication and multi-tenant architecture
• Dashboard UI for viewing parsed email data

Week 3: Polish & Deploy

• Queue-based architecture for handling email processing at scale
• AWS deployment with proper monitoring
• Final QA and investor demo preparation

What We Built

The MVP included:

• Secure email connection (OAuth for Gmail/Outlook)
• AI-powered parsing that extracts structured data from unstructured emails
• Real-time dashboard showing parsed information
• Export functionality for validated data
• Admin panel for monitoring processing status

The Tech Decisions

Why Claude over GPT-4? At the time, Claude's structured output capabilities were more reliable for our parsing use case. We also built an abstraction layer that would allow swapping models if needed.

Why Queue-based? Email parsing can be slow and unpredictable. A queue-based architecture meant the UI stayed responsive while processing happened in the background—critical for a good demo experience.

Why PostgreSQL over NoSQL? The data was highly relational (users → emails → parsed fields). PostgreSQL's JSON columns gave us flexibility for unstructured data while maintaining referential integrity.

Results

The demo went flawlessly. Investors saw a real product processing real emails in real-time. Cubby closed their pre-seed round shortly after, with the MVP serving as the foundation for continued development.