✨ Inspiration
We’ve both done research and built products where recruiting participants was a massive bottleneck. Cold emails, long spreadsheets, and messy scheduling made it hard to focus on the actual work. With the rise of autonomous AI agents, we saw a chance to rethink the entire participant recruitment pipeline — and LinkLine was born.
⚙️ What it does
LinkLine is an AI agent system that automates participant recruitment for research and product testing.
It:
- Takes in a research brief + inclusion/exclusion criteria
- Uses Exa to find potential participants (email, LinkedIn, phone)
- Uses CrewAI agents with Gemini to generate personalized outreach emails
- Uses CrewAI + Vapi to orchestrate and execute real-time phone calls to participants
The end result is a system that helps researchers recruit faster, smarter, and with far less manual effort.
🛠️ How we built it
- We built our backend using Flask and stored research sessions in SQLite
- Used Exa to search the public web and extract contact info for potential participants
- Used Google Gemini as the reasoning engine behind our CrewAI agents
- Built email outreach flows powered by CrewAI agents using LLM-generated messaging
- Used CrewAI + Vapi to make automated voice calls driven by the research criteria
- Created a simple HTML + JS frontend to collect researcher input and trigger agent workflows
🧱 Challenges we ran into
- Figuring out authentication for sending emails reliably
- Getting multi-agent systems to work with multiple tools (CrewAI, Vapi, Gemini)
- Parsing and filtering Exa’s output to match inclusion/exclusion criteria
- Limited or broken documentation for some of the APIs we integrated
🏆 Accomplishments that we're proud of
- Built a fully functional pipeline: criteria input → participant discovery → outreach via email and phone
- Integrated multiple platforms (Exa, CrewAI, Vapi, Gemini) into one seamless system
- Got real-time phone calls working with dynamic, criteria-based call logic
- Successfully used Exa despite its broken or unclear documentation
- Delivered an end-to-end solution that mirrors what real-world researchers and builders need
📚 What we learned
- Agent orchestration tools like CrewAI are powerful with the right task breakdown and prompt design
- Vapi voice agents unlock a whole new layer of communication when paired with good LLM logic
- Exa is incredibly useful but requires careful post-processing to make it usable
- Working with multiple AI tools in one pipeline is hard — but totally worth it when it clicks
🚀 What's next for LinkLine
- Automate scheduling and confirmations using Calendly API
- Add support for new outreach channels like LinkedIn, Discord, and SMS
- Build a dashboard for researchers to track outreach and engagement
- Launch LinkLine as a real tool for researchers, founders, and user testing teams
Built With
- crewai
- exa
- flask
- google-gemini
- html
- javascript
- python
- vapi
- wandb
- weave
Log in or sign up for Devpost to join the conversation.