Inspiration
Many promising ideas begin as a few unclear sentences. Existing AI tools often respond with an immediate answer, but they rarely help users discover missing assumptions, examine evidence, resolve contradictions, and gradually develop the idea into a practical plan.
ROAST was inspired by the belief that early ideas need thoughtful collaboration, not instant judgment.
What it does
ROAST is an AI-powered idea development workspace that turns a vague idea into a structured, evidence-backed, and actionable plan.
It guides users through clarifying the problem, identifying the target audience, gathering relevant evidence, exploring the idea from multiple perspectives, evaluating risks, and designing low-cost validation experiments.
GPT-5.6 acts as the lead reasoning layer, coordinating the process and synthesizing the results into a clear proposal that users can continue refining.
How we built it
ROAST was built with React, TypeScript, Vite, Node.js, SQLite, and server-sent events for real-time AI responses.
The application includes a multi-model provider layer, an evidence-gathering system, structured discussion workflows, persistent project history, and document export tools.
During Build Week, Codex was used to redesign and rebuild the core idea-to-plan workflow, iterate on the product experience, implement integrations, and improve the application through continuous testing. GPT-5.6 powers the main clarification and synthesis process inside the product.
Challenges we ran into
The main challenge was preventing the experience from becoming an unstructured conversation between several AI models.
We needed to give every AI perspective a clear purpose, keep the user involved, distinguish evidence from assumptions, and ensure that the final result was more useful than a long transcript.
Another challenge was designing a workflow that provides structure without forcing every idea through the same rigid template.
Accomplishments that we're proud of
We created a working end-to-end workspace that transforms an early idea into a practical proposal instead of producing another one-shot AI answer.
We are especially proud of the evidence-aware analysis, the user’s ability to participate throughout the process, and the final output that connects assumptions and risks with concrete validation steps.
What we learned
We learned that developing an idea is not a single generation task. It is an iterative process of clarification, research, challenge, synthesis, and revision.
We also learned that multiple AI perspectives are most valuable when they operate inside a clearly designed workflow and contribute toward a shared outcome.
What's next for ROAST — Idea-to-Plan Workspace
Next, we plan to add reusable project workspaces, deeper evidence gathering, collaborative editing, progress tracking, and validation results that can update the original proposal.
Our long-term goal is to make ROAST a continuous workspace where an idea can evolve from an early thought into a validated product or initiative.
Log in or sign up for Devpost to join the conversation.