Socratic

Socratic is an adaptive tutor that helps students think through homework without handing them copyable answers.

Students bring a problem in math, science, writing, or logic. Instead of solving it, Socratic asks one targeted question at a time, detects misconceptions, calibrates hints, and keeps the student moving through their own reasoning.

What it does

Guides students through a focused Socratic dialogue rather than giving solutions.
Detects misconceptions and adapts the next question to the smallest useful step.
Responds warmly to stuckness and direct-answer requests without leaking the answer.
Declines active exams, tests, and quizzes while offering practice afterward.
Maintains a live Tutor's Notebook with progress, confidence, hint depth, frustration, misconceptions, and mastery state.
Adds a fresh analogous Mastery Check after the original problem; progress reaches 100% only when the student independently transfers the idea.
Produces an anonymous, shareable session report and private on-device progress history.

Why it matters

Most answer machines optimize for finishing homework. Socratic optimizes for the moment a learner can explain why their approach works. That makes progress more honest for students, clearer for parents or teachers, and more useful than an answer alone.

How we built it

Socratic is a Next.js application with a structured tutoring state contract, streaming model responses, server-side guardrails, local session persistence, and a responsive learner workspace.

We used Codex with GPT-5.6 throughout the build to turn the tutoring plan into working product code: designing the state model and prompt contract, implementing safety and answer-leak protections, building the adaptive UI, diagnosing edge cases from evaluation reports, and repeatedly testing and refining the deployed experience.

The production tutor runs through Fireworks AI and is deployed on Vercel.