All Case Studies
Case Study

EdTech MVP: AI-Powered Tutoring Platform for K-12

An AI tutoring platform that generates personalized practice problems, provides step-by-step explanations, and tracks student progress across math and science.

Client: BrainSpark Education

Timeline
14 days
Investment
$7,499
Key Result
2,400 students, 85% weekly retention

Student view showing a math problem with an AI chat sidebar giving step-by-step hints, a progress bar showing topic mastery, and a streak counter. Clean, colorful UI designed for teenagers.

The Challenge

BrainSpark's founder was a former math teacher who watched students struggle with one-size-fits-all worksheets. Some students needed harder problems, others needed more scaffolding, and nobody got personalized feedback. Khan Academy was too passive (videos), and existing AI tutors just gave answers without teaching the reasoning. She wanted a platform that adapted difficulty in real-time, used the Socratic method (guiding questions instead of answers), and gave teachers visibility into where each student was stuck. She had a $10,000 budget and a school district willing to pilot with 500 students if the tool was ready by September.

Our Approach

The core innovation was the AI tutor personality. We prompt-engineered Claude to act as a Socratic tutor: never giving answers directly, instead asking guiding questions that lead students to discover the solution themselves. When a student got stuck, the AI would break the problem into smaller steps and ask 'What do you think the first step would be?' The adaptive engine tracked mastery per topic (fractions, algebra, geometry, etc.) and adjusted problem difficulty using a simple Elo-like rating system. Problems were generated dynamically by Claude based on the student's current level and weak areas. The teacher dashboard showed class-wide analytics: which topics had the most struggling students, individual progress curves, and flagged students who hadn't logged in. We kept the student UI deliberately simple and colorful with streak counters and achievement badges to drive engagement. The architecture was Next.js with PostgreSQL, and we used streaming responses for the AI tutor to feel conversational.

What We Built

Socratic AI tutor with streaming conversational interface.
Adaptive problem generator that adjusts difficulty per student per topic.
Student dashboard with mastery tracking, streaks, and achievements.
Teacher dashboard with class analytics and struggling-student alerts.
Parent view with weekly progress summaries sent via email.

Delivery Timeline

Day 1-3: Foundation

Auth with class codes, database schema for students/teachers/progress, AI tutor prompt engineering.

Day 4-7: AI Tutor Core

Streaming chat interface, Socratic prompt system, adaptive problem generation, mastery tracking.

Day 8-10: Student Experience

Dashboard, streaks, achievements, topic selection, practice sessions.

Day 11-12: Teacher Dashboard

Class analytics, struggling-student alerts, topic heat maps, individual progress.

Day 13: Parent View + Email

Weekly progress digest, parent dashboard, Resend email integration.

Day 14: Launch

Production deployment, seed content for math/science, school pilot setup.

Tech Stack

Next.js
Frontend + API
Claude AI
AI Tutor
PostgreSQL
Database
Drizzle ORM
ORM
Resend
Email
Redis
Session Cache
Railway
Hosting
Vercel
Frontend

Architecture

frontend

Next.js with a kid-friendly UI using Tailwind CSS custom theme.

backend

Next.js API routes with streaming for AI responses. Drizzle ORM.

auth

Better Auth with class codes for students, email for teachers/parents.

data

PostgreSQL for user data and progress tracking. Redis for session state.

ai

Claude 3.5 Sonnet with custom Socratic tutor system prompt and streaming.

Security

coppa

COPPA-aware design: no personal data collected from under-13 without parent consent.

rbac

Student, Teacher, Parent, Admin roles with strict data boundaries.

monitoring

AI response monitoring for safety. Content filters on all outputs.

qa

Automated tests for problem generation accuracy and AI safety filters.

The Results

Students enrolled (month 1)
02,400
Weekly retention rate
N/A85%
Average mastery improvement
Baseline+23% in 4 weeks
My students actually ask to use BrainSpark during free time. The AI tutor feels like talking to a patient teacher, not a robot. One kid went from failing fractions to scoring 92% in three weeks.
Michael Rodriguez
8th Grade Math Teacher

Key Takeaways

The Socratic method works brilliantly with LLMs. Prompt Claude to never give answers, only ask guiding questions. Students learn more and engage longer.

Streaks and achievements drive daily usage in K-12. BrainSpark's 85% weekly retention is almost entirely driven by the streak counter.

Teacher buy-in is the real product. Students use what teachers assign. The class analytics dashboard was what convinced the district to pilot.

Deliverables

Full source codeAI tutor prompt libraryTeacher onboarding guideStudent analytics pipelineProduction deployment

FAQ

Frequently Asked Questions

Related Case Studies

Want similar results?

Book a free 15-min scope review. Your vision, engineered for production in 14 days. Fixed price.

Book Scope Review