What it does
How we built it
Sofon is an AI research companion that turns informal scientific intuition into a precise, testable hypothesis. Many scientific ideas begin as metaphors or surprising connections, but the hard part is separating evidence, inference, speculation, and metaphor. Sofon helps by transforming an ordinary-language idea into a clear claim, supported knowledge, hidden assumptions, falsifiable predictions, a minimal discriminating experiment, red flags, and the single best next question. It always includes what result would support the idea and what result would argue against it. Built as a Streamlit app using GPT-5.6 through the OpenAI Responses API, Sofon uses a structured prompt to enforce epistemic separation and render the result as readable scientific cards. The biggest challenge was preventing the AI from becoming either a cheerleader or a generic critic. We focused the output on disconfirmation, minimal tests, and explicit uncertainty. Sofon is useful as both an education tool and an early-stage research ideation companion: it does not decide whether an idea is true, but makes it precise enough to be challenged.
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Sofon: From Scientific Intuition to Testable Hypothesis
Built With
- codex
- openai-gpt-5.6
- python
Log in or sign up for Devpost to join the conversation.