Inspiration
Screen-time management often becomes a daily negotiation between parents and children.
Parents want safety, boundaries, and visibility. Children want fairness, independence, and the opportunity to earn trust. Most existing tools focus primarily on restriction: block an application, lock a device, or display a usage report. They help control access, but they do not necessarily help children develop healthier digital habits.
We created ScreenSmart Family OS around a different idea:
Screen time should not only be restricted. It should help children learn responsibility, build healthy habits, and gradually earn autonomy.
ScreenSmart combines parental controls with learning challenges, family agreements, rewards, and a Screen Bank where screen time becomes an age-appropriate learning currency.
What it does
ScreenSmart is a family digital-wellbeing platform that connects the parent experience, child experience, policy engine, learning system, and privacy layer.
Parents can:
- Set daily screen-time limits and scheduled routines
- Manage application access and approvals
- Review device and protection status
- Create learning, habit, safety, or responsibility goals
- Review uncertain AI assessments before rewards are granted
- Track progress without reading private conversations or using hidden surveillance
Children can:
- See their remaining screen time clearly
- Complete age-appropriate learning or responsibility challenges
- Earn additional screen time through positive actions
- Save unused minutes in their Screen Bank
- Learn concepts such as saving, interest, borrowing, repayment, and trade-offs using time instead of real money
- Participate in parent-child reward agreements
- Build loyalty points, streaks, and progress over time
The objective is to replace repeated parent-child conflict with a transparent system of rules, effort, progress, and earned trust.
Our OpenAI Build Week focus
ScreenSmart is a broader product that existed before OpenAI Build Week. We are not presenting the entire historical platform as work completed during the hackathon.
For Build Week, we are focusing on a clearly bounded, demonstrable workflow:
- A parent selects a learning, digital-safety, or healthy-habit goal.
- ScreenSmart creates an age-appropriate challenge.
- The child completes the challenge.
- The submission is evaluated against a structured rubric.
- High-confidence results can be processed automatically.
- Uncertain results are routed to the parent instead of allowing an unsafe or unreliable automated decision.
- Successful completion grants a controlled reward such as bonus screen time, Screen Bank credit, or loyalty progress.
- The parent receives a concise outcome and audit trail.
This creates a complete loop from parental intent to child action, assessment, human oversight, and earned autonomy.
Before final submission, this section will be updated with the exact features, commits, tests, and Codex sessions completed during the Build Week eligibility period.
How we are using OpenAI Codex
We are using OpenAI Codex as an engineering agent rather than treating it as a generic code generator.
Codex is helping us:
- Read the existing product requirements and repository before changing code
- Trace product requirements to database, API, frontend, and test evidence
- Identify incomplete or conflicting implementation paths
- Plan minimal changes instead of broad rewrites
- Implement bounded features across the relevant modules
- Generate and improve regression tests
- Review authentication, authorization, privacy, and child-safety boundaries
- Run type checking, tests, linting, and build validation
- Document changed files, assumptions, risks, and remaining limitations
The human role remains essential. Product boundaries, child-safety rules, privacy decisions, architecture, and final acceptance are reviewed and controlled by the project owner.
How we built it
ScreenSmart uses a modular monorepo architecture.
The platform includes:
- Next.js and TypeScript for the parent, school, and administrative web experiences
- Fastify and TypeScript for backend APIs
- PostgreSQL and Prisma for transactional data and auditable ledgers
- Redis for caching, rate limiting, and operational workflows
- Firebase Authentication for account identity
- Kotlin for the Android child agent
- .NET for the Windows child agent
- Chrome and Edge Manifest V3 for browser-level controls
- Google Cloud infrastructure for deployment, storage, secrets, and operational services
The Build Week demonstration is intentionally focused on the web/API and supported non-Apple product paths. Apple child-device enforcement remains a later platform phase and is not presented as completed functionality.
Privacy and child safety
A child-focused product must be designed differently from a normal consumer application.
ScreenSmart is based on the following boundaries:
- No child-targeted advertising
- No sale of child data
- No private-message or chat scraping
- No keylogging
- No hidden screenshots
- No covert monitoring
- Parental consent before processing child data
- Data minimisation and purpose limitation
- Human review when an automated assessment is uncertain
- Honest disclosure where an operating system does not permit direct enforcement
The parent should receive actionable information, not a surveillance feed.
Challenges we faced
Balancing control with autonomy
A parental-control product can easily become overly restrictive. The challenge was to create enforceable boundaries while still giving the child a fair and motivating path to earn additional freedom.
Designing safe automated assessment
AI output should not directly determine a child’s access when confidence is low. We therefore designed confidence thresholds, structured outputs, safe failure modes, and parent-review fallbacks.
Coordinating multiple product surfaces
A single challenge may involve the parent interface, child interface, API, database, reward ledger, entitlements, notifications, and audit history. Maintaining consistency across these layers required explicit contracts and requirement-to-code traceability.
Remaining honest about platform limitations
Android, Windows, browsers, and Apple platforms expose different capabilities. ScreenSmart must not claim enforcement that an operating system does not technically permit.
What we are proud of
We are proud that ScreenSmart approaches digital wellbeing as a behaviour-design and family-trust problem rather than only a device-blocking problem.
The combination of:
- parental controls,
- earned learning,
- Screen Bank,
- parent review,
- reward contracts,
- privacy controls, and
- auditable engineering
creates a more constructive relationship between children, parents, and technology.
What we learned
The most important lesson is that AI-assisted engineering becomes significantly more useful when the AI agent is given:
- an explicit source of truth,
- strict architectural boundaries,
- evidence requirements,
- validation commands,
- safety constraints, and
- a definition of done.
Codex is most valuable when it works inside a disciplined engineering process where every claim can be connected to implementation and test evidence.
What's next
Our next priorities are:
- Complete and validate the Build Week end-to-end demonstration.
- Publish the exact pre-existing-versus-Build-Week change record.
- Expand physical-device testing for supported platforms.
- Strengthen school and parent pilot workflows.
- Conduct structured security, privacy, and child-safety reviews.
- Continue developing Apple platform support as a separately validated future phase.
Our long-term goal is to make ScreenSmart a trusted Family OS where screen time grows with the child—not simply a switch that turns their device on or off.
Built With
- .net
- api
- authentication
- cloud
- codex
- extension
- gemini
- github
- kotlin
- next.js
- node.js
- platform
- postgresql
- react
- redis
- turborepo
- typescript
Log in or sign up for Devpost to join the conversation.