FundWave

Inspiration

I’ve been using Claude Code with a custom frontend for my fundraising. It worked fine, but I basically rebuilt Airtable. I wanted something more agentic—something that could actually help me, not just hold data.

What it does

FundWave connects shared investor data with AI agents and a chat interface. It handles research, scoring, and follow-ups, and it’s meant to get smarter and more independent—especially once the calendar and research integrations are fully working.

How we built it

I used Claude Code, Roo Code, and ChatGPT—often all at once—plus sponsor tech like Mastra, CoreWeave, CopilotKit, and Tavily. The goal was to give each framework a fair shot while wiring them into one workflow.

Challenges

The main challenge was making it actually agentic, not just another chatbot. Getting it to take initiative—look ahead, schedule things, follow up—takes more than prompt tuning. Also, Claude likes to ignore instructions and just “fake” progress; it even wrote its own orchestrator instead of using Mastra.

Accomplishments

It now writes solid follow-up emails in my voice and pulls from a shared glossary and docs, so tone and context stay consistent.

What we learned

AI agents are still unreliable coworkers. They’ll claim they did the job, but they’ll quietly skip the parts they didn’t understand. Transparency and verification are everything.

What’s next

Finish the calendar integration, speed up transcript drag-and-drop, and push it toward real proactivity—something that can look ahead and help without being asked.

Built With

• ag-ui
• google-cloud
• mastra
• openai-api
• python
• tavily
• typescript