StudyMate

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for StudyMate

Inspiration

Working adults who study alongside a full-time job do not simply need another to-do list. They need to decide what to study, how much time each subject deserves, and how to adjust when real life disrupts the plan. StudyMate grew from that firsthand problem: making limited weekday and weekend study time realistic, actionable, and sustainable.

What it does

StudyMate converts subjects, priorities, a target period, and weekday/weekend availability into a hierarchical monthly, weekly, and daily plan. A learner completes an item, records the topic, duration, and understanding level, receives AI-generated next actions, and sees the result reflected in history, statistics, and the dashboard.

How we built it

StudyMate uses the OpenAI Responses API with GPT-5.6 Terra. GPT-5.6 designs monthly and weekly goals, creates executable daily items within time limits, and analyzes completed study sessions to return structured feedback and concrete follow-up actions. Structured Outputs are backed by Zod schemas. The server validates dates, subject assignments, and daily capacity before generated data is persisted.

Codex accelerated the full delivery loop: architecture, the plan-generation pipeline, Supabase social authentication, PostgreSQL data modeling, responsive UI, unit and browser tests, production debugging, Vercel deployment, and submission assets. Human-directed decisions included the target user, the plan → action → record → feedback loop, privacy boundaries, and removing unreliable UI rather than presenting broken features.

Challenges

The largest challenge was combining generative planning with strict calendar constraints. StudyMate first builds deterministic month and week boundaries, asks GPT-5.6 for structured content, validates every generated item, and retries invalid output with the validation error as context.

Accomplishments

• A complete responsive product rather than a single-prompt demo
• Multi-provider authentication
• Hierarchical planning with regeneratable weekly schedules
• Study records linked directly to planned work
• Personalized GPT-5.6 feedback and actionable next steps
• Progress dashboards and subject-level statistics
• Korean and English localization

What we learned

AI planning is most useful when the model receives explicit constraints and its output is validated like any other untrusted external input. The strongest experience combines model reasoning with deterministic application rules.

What's next

Calendar-aware rescheduling, retrieval from course materials, spaced-repetition recommendations, and collaborative views for mentors and study groups.

Built With

• openai
• responses