AI Assistant MVP Cuts Founder Admin Time 78%
SpeedMVPs built VeloraAssist's AI virtual assistant MVP in 3 weeks: an AI EA that reads email context, suggests replies and books meetings, running on a Gmail/Outlook OAuth pipeline, a thread-aware Claude-backed reply engine, structured tool calls for calendar actions and a Next.js dashboard. Eval-driven prompts let accuracy improvements ship without regressions, and the assistant cut founder admin time 78%.
A 3-week MVP combining a Gmail/Outlook OAuth pipeline, a thread-aware Claude-backed reply engine, structured tool calls for calendar actions, and a Next.js dashboard. Built with eval-driven prompts so accuracy improvements ship without regressions.
Book a free 30-minute callProductivity SaaS
Industry
3 Weeks
Timeline
Project Overview
1
What We Built
- A 3-week MVP combining a Gmail/Outlook OAuth pipeline, a thread-aware Claude-backed reply engine, structured tool calls for calendar actions, and a Next.js dashboard. Built with eval-driven prompts so accuracy improvements ship without regressions.
2
The Challenge
- VeloraAssist's founder had eight pilot customers who all wanted the same thing, an AI EA that could read email context, suggest replies, and book meetings, but every off-the-shelf tool either lacked context or leaked it. They had three weeks before a YC interview and no time for a 90-day build.
3
Our Solution
- Thread-aware reply drafting with citation-style context highlighting
- Tool-calling agent that books, reschedules, and declines with calendar awareness
- Per-tenant prompt overrides so each founder's voice stays unique
- Eval suite with 240 golden conversations regression-tested on every prompt change
- PII redaction and zero-retention provider configuration for GDPR pilots