Parent-Co-Teacher

• 

  App screen showing supplementary content generated by the app for the parent

• 

  Quiz to make sure that kids have grasped the concept

• 

  Supplementary Material Generation on the fly to solidify the learning

• 

  Questions are explained in native language to simplify the learning

The Problem

In India, 250 million+ parents struggle to help their children with homework—not because they don't care, but because their children attend English-medium schools while they themselves have limited English literacy.

Current reality:

• Child brings home English homework
• Parent can't read or understand the content
• Child's learning depends on parent's language skills, not their dedication
• Educational inequity perpetuates across generations

Our Solution

Parent-Co-Teacher transforms homework images into actionable parent-child teaching scripts in Hinglish (natural Hindi-English code-mixing), with AI-powered audio so parents can teach effectively regardless of their own literacy level.

Key Features

Marathon Agent: Autonomous Planning

Our standout feature is an autonomous learning planner that:

1. Analyzes session history and weakness trends
2. Generates a personalized 7-day learning plan
3. Self-critiques using quality metrics (50-100 scale)
4. Refines the plan before presenting to parents
5. Adapts missions based on completed work via check-ins

This isn't a chatbot—it's a stateful orchestration system that learns and improves.

Tech Stack

• Frontend: React 19 + TypeScript + Tailwind CSS + Vite
• AI: Gemini 3 Pro (Vision + Reasoning) + Gemini 3 Flash (Generation) + Gemini TTS
• Storage: IndexedDB (client-side, privacy-preserving)

Challenges we ran into

1. Reliability for live demo with mixed scan quality and varied textbook layouts
   
2. Natural language quality (avoiding literal/robotic Hindi translation)
   
3. Consistency across workflows (homework, chapter, quiz, micro-lesson, marathon)
   
4. Stateful autonomy (moving from one-shot answer generation to long-running planning)

Accomplishments that we're proud of

1. Accessibility Drives Innovation: Designing for parents who can't read English forced us to build better UX for everyone. Audio-first, visual-guided interfaces work better than text-heavy alternatives.

2. Self-Critique Requires Structure: Asking an LLM to "improve this" doesn't work. Providing a concrete verification checklist with specific criteria produces measurable improvements.

3. Stateful AI > Stateless Prompts: The power of our Marathon Agent comes from memory—session history, weakness trends, mission progress. Single-prompt chatbots can't achieve this.

4. Code-Mixed Language is Hard: Hinglish isn't Hindi or English—it's a dynamic blend that varies by region and context. We learned to specify "natural Mumbai Hinglish" for consistent, authentic output.

5. Real Users Ground Truth: Every design decision traces back to "Would Sunita understand this?" This persona-driven development kept us focused on actual impact over technical elegance.

What's next for Parent-Co-Teacher

1. Voice Input: Let parents ask questions by speaking instead of typing. "Yeh question ka answer kya hai?" triggers voice-to-text and generates a response.

2. Regional Language Expansion: Start with Tamil-English, Telugu-English, and Marathi-English code-mixing for South and West India.

3. Teacher Dashboard: Let teachers assign homework through the app and track which parents engaged with it—bridging the school-home gap.

4. WhatsApp Integration: Most Indian parents live on WhatsApp. A bot that accepts homework photos and returns audio guidance would dramatically increase accessibility.

5. Community Curriculum: Let successful parents share their Marathon Agent plans with others facing similar challenges. Crowdsourced, proven learning paths.

6. Offline Mode: Full offline functionality for areas with unreliable internet. Pre-download curriculum content and sync when connected.

Built With

• gemini
• indexdb
• react
• tailwind
• typescript
• vite