Filling out forms is often a tedious process whether it’s job applications, hackathon registrations, or even feedback surveys. Inspired by this constant friction, I built Formyr, a conversational form builder and filler powered by an AI agent. Instead of manually typing into dozens of input fields, users simply chat with the agent, which fills the form step by step and knows when the submission is complete.
I was inspired by how repetitive and frustrating traditional forms can be and wanted to make the process feel more natural, like talking to someone who already knows what’s needed. I built the project with Next.js for the frontend, Postgres with Drizzle ORM for storage, Clerk for authentication, and LLMs connected via tool calling to parse responses, update draft submissions, and generate structured JSON output.
Through this project I learned how to design reliable AI agents that manage multi-step conversations, how to normalize free-text input into structured schema fields, and how to balance public access with secure endpoints. One of the main challenges was getting the model to consistently ask one question at a time without drifting, as well as ensuring partially filled forms could be saved as drafts so users never lose progress. Working against the clock during the hackathon also made deployment, validation, and analytics design especially challenging, but ultimately it pushed me to ship an end-to-end product that turns form filling into a simple conversation.
Log in or sign up for Devpost to join the conversation.