Donna

Dashboard - With calendar and Donna
Calander Section - Donna has access
Profile - Donna knows some info about you
Call History/Transcript of Donna's Call
Settings - Extra

Inspiration

Everyone important has an assistant who handles the small things. But most people don’t. We were inspired by the idea that while businesses have automation, consumers still spend time making phone calls just to check availability, get pricing, or book appointments. Inspired by Donna Paulsen from Suits, the person who always “just handles it,” we wanted to build something that feels like having your own personal executive assistant. The goal was simple: make an AI that gets things done.

What it does

Donna is a personal AI assistant that handles real-world tasks for you. You tell Donna what you need — like finding and booking a salon appointment — and she researches the best options online. If the information isn’t available on the internet, Donna calls the business herself, asks the right questions, and handles the booking. If the business is closed, she schedules the call for when they open and updates you when it’s done. Once an appointment is confirmed, Donna automatically adds it to your Google Calendar so your schedule stays organized.

How we built it

Donna is powered by Gemini as the reasoning engine that decides what actions to take. We use Tavily for web search to gather information about businesses. When information isn’t available online, Vapi combined with ElevenLabs voice allows Donna to place natural-sounding outbound phone calls. Call transcripts are stored in MongoDB, where Gemini summarizes results and extracts booking details. The frontend is built with Next.js, while Node.js and cron jobs handle scheduled calls when businesses are closed. The system integrates with Google Calendar API integration so confirmed bookings automatically appear in the user’s schedule.

Challenges we ran into

One challenge was bridging online information with real-world phone calls. We had to design a reliable decision loop where Donna first searches the web, then decides whether a phone call is necessary. We also had to ensure calls were legally compliant by including disclosure that the caller is an AI assistant, especially in two-party consent states like Florida.

Accomplishments that we're proud of

We’re proud that Donna doesn’t just generate answers, she takes real action. She handles appointments, restaurant reservations, service inquiries, price comparisons, and many other everyday tasks that currently require phone calls. The system works end-to-end: searching for businesses, calling them, extracting information from the conversation, and adding them to a calendar automatically. Her calls are time efficient, we gather call transcripts, and have full calendar integration and profile customization. Just like in Donna Paulsen from Suits, Donna knows everything.

What we learned

We learned that building an autonomous AI agent requires more than just a language model. It requires orchestration between multiple systems: search APIs, voice agents, databases, schedulers, and user interfaces. We also learned the importance of designing clear decision logic so the AI knows when to search, when to call, and when to act on behalf of the user.

What's next for Donna

Next, we want to expand Donna’s capabilities to work with multiple apps. For example, telling Donna to cold message 20 Tech companies on Linkedin. We also want to improve memory and personalization so Donna learns user preferences over time and proactively manages their schedule. Ultimately, the goal is to turn Donna into a true executive assistant that manages all things online.