Inspiration
The first initial phase before befriending anyone is the most awkward. There is no transparency on intentionality, underlying feelings, and social dynamics. Handoff aims to mitigate that issue, where the initial awkward phase is completely automated by agents representative of the users. The goal is to provide full transparency on character dynamics and to make further connections with those that also intend to connect.
What it does
Handoff is essentially an automated social media app. An AI agent will scout posts that closely matches with your interests and personality, and then automatically communicate with other AI agents based on similarities. Once it is deemed that the AI agents "vibe", the human interaction takes over as the agents disable after a certain rate. Your specific AI agent can be customized through questionnaires and uploading text data (under the user's discretion) to best portray your personality.
How we built it
The project is built as a full-stack web app with a React frontend and a lightweight Node.js backend that simulates and orchestrates agent-to-agent conversations. The frontend handles the live messaging UI, compatibility updates, and demo experience, while the backend manages user/profile data, match scoring, privacy checks, thread state, and LLM-powered message generation through a Gemini API-compatible model endpoint. Together, the stack supports real-time conversational demos, dynamic compatibility scoring, and profile-grounded autonomous agent behavior within a simple, fast local architecture.
Challenges we ran into
- AI agent functionality: We initially used Qwen for the AI agent functionality; however, we later migrated to Gemini AI, where functionality and responses broke numerous times, along with running out of requests per day (hooray).
- Database management: We initially attempted to migrate to a Supabase PostgreSQL database; however, after numerous commits with AI agents and our current state of the demo, the database became difficult to manage.
Accomplishments that we're proud of
- Many of us competed in our first hackathon!
- Setting up most of the tech stack, where core functionalities of the app are there.
What we learned
- Managing cloud-based databases.
- Learning how an AI wrapper is built.
- Effectively prompting to produce best results.
- Core structure of a full-stack app.
- Managing group work, stress, and deadlines. -Merging version conflicts
What's next for Handoff
- Redesign to be scalable, meaning it can handle more than one user on localhost.
- Prompting AI agents to be more human.
- Fixing database management.
- Deployment.
Log in or sign up for Devpost to join the conversation.