Please read as it covers technical details not mentioned in video
Inspiration
Often, we have found ourselves and our friends wanting to organize community volunteering events to improve their neighborhoods and beyond. However, we have noticed there is often significant planning that needs to be done with deciding what resources are needed, finding and allocating volunteers, in coordinating these events.
What it does
Mastermind solves all of these issues by taking basic inputs such an idea for a social project, and generates an in-depth flowchart including information like materials required, breaking up a complex task into easier steps, and even assigning volunteers to these steps. Volunteers are assigned based on a "reliability score," which is an aggregate of participation in their past projects, and sentiment analysis of their chat comments to get a holistic overview of which volunteers would be most suited for what tasks. The app also provides real-time chat for collaboration, automatically adjusts assignments if volunteers become unavailable, and predicts progress and resource usage, ensuring projects run efficiently from start to finish.
How we built it
For logistics planning, we used the Cohere LLM and extended it with agentic capabilities, creating a custom multi-step prompt that combines user input with historical project data, volunteer availability, and resource constraints.
The LLM programmatically generates a dynamic, easy-to-read flowchart via React Flow, breaking down complex projects into actionable steps and assigning volunteers to tasks based on an aggregate productivity score.
This score integrates past participation rates, task completion rates, and a real-time sentiment analysis of chat interactions, giving a holistic measure of each volunteer’s reliability, engagement, and suitability for specific tasks.
By combining AI-driven task decomposition, adaptive volunteer allocation, and predictive insights, Mastermind transforms unstructured project ideas into structured, executable social impact plans that optimize efficiency, engagement, and collaboration.
Challenges we ran into
One of the main challenges we ran into was dynamically creating the flowchart. We had to focus on the architecture of the graph, and how edges and nodes would connect and be organized to generate a visual that balances comprehensive and intuitive.
We also spend a lot of time linking various tables across Supabase, the database we used for storing the projects, chats, and the volunteer database.
Accomplishments that we're proud of
We're proud of the fact that we could design an AI Agent that can complete tasks in an advanced, but also intuitive way. Learning a little about graphs (ex. edges and nodes) was also new and interesting for us in how it enables data to be visualized uniquely. This was also the first time we implemented a live chat, so it was interesting to learn about how fast inference with databases is necessary.
What we learned
Through building Mastermind, we gained a deeper understanding of integrating AI with real-world workflows. We learned how to design prompts for an LLM that balance flexibility with structure, ensuring actionable output. Implementing the flowchart also taught us about graph theory in practice. We also discovered the importance of combining multiple data signals like past participation, task completion, and chat sentiment, to create a reliable volunteer assignment system. Finally, building a real-time collaborative chat highlighted the challenges of syncing database state efficiently, which taught us the value of careful backend design for nice user experiences.
What's next for Mastermind
Next, we hope to add more in-the-moment features, where volunteers can log progress as they volunteer. We also think it would be really useful to eventually add a feature where volunteers can upload photos and other media and we can use multimodal generative AI to create flyers and social media posts. This would attract more volunteers and supporters, and bring even more effectiveness to the app.
Log in or sign up for Devpost to join the conversation.