Inspiration

Traditional exam preparation often falls into the trap of rote memorization. Students often feel lost in a "generic chatbot" experience that either gives the answer too quickly or fails to explain the underlying logic. We wanted to build a tutor that passes the "First 3 Seconds" test: an interface so engaging and transparent that students immediately trust it as their personal pedagogical guide for the UTBK and beyond.

What it does

Eduvoka is an AI-powered learning platform that transforms typical quiz-taking into a Deep Inquiry experience.

  • Socratic AI Tutor: Instead of just revealing answers, our Gemini-powered tutor guides students through four pedagogical phases: Ask, Clarify, Challenge, and Conclude.
  • Pedagogical Transparency: We replaced the "black box" of AI with Thinking Visualizations, Phase Indicators, and a Live Concept Map that builds a knowledge graph as the student learns.
  • Learning Velocity Analytics: Features a specialized dashboard that calculates DTA (Dynamic Thinking Agility) vs MBA (Mastery Behavioral Analysis) to identify whether a student's struggle is a technical gap or cognitive burnout.
  • Behavioral Mapping: A unique 2D visualization that clusters student performance to provide hyper-personalized recommendations.

How we built it

  • Framework: Built with Next.js and TypeScript for a robust, type-safe full-stack architecture.
  • AI Core: Powered by Google Gemini (gemini-1.5-flash), utilizing custom system instructions for strict Socratic pedagogical adherence.
  • Database: Prisma ORM with PostgreSQL, managing complex relations between users, materials, and practice sessions.
  • Visualization: Leveraged Tailwind CSS and Lucide React for premium, high-fidelity UI components including animated thinking steps and knowledge graph nodes.
  • State Management: React hooks for real-time conversation flow and analytics persistence.

Challenges we ran into

  • AI Infinite Loops: Early versions of the Socratic tutor would explore tangents endlessly. We implemented a Strict 8-Turn Inquiry Engine that automatically triggers a "Mastery Evaluation" phase to ensure closure.
  • Database Resilience: Handling unique constraints during high-speed database seeding for 75+ complex questions required refactoring our Prisma upsert logic.
  • Pedagogical Flow: Ensuring students didn't accidentally "skip to the result" without undergoing the Socratic intervention required a complete overhaul of our Next.js redirection logic and bridge pages.
  • Metric Sensitivity: Defining "Burnout" thresholds was tricky. We adjusted our mathematical models to provide valid analysis even for short, high-intensity study sessions (lowering the data threshold from 5 to 3 questions).

Accomplishments that we're proud of

  • Pedagogical Integrity: Building a tutor that actually "refuses" to give the answer, instead leading the student to their own "Aha!" moment.
  • The "Thinking" UI: Creating a visual language for AI reasoning that makes the student feel like they are working with a mentor, not just a machine.
  • Rapid Localization: Successfully localizing a massive 1500-line database seeder and the entire dashboard UI from Indonesian to English to ensure a global-ready hackathon entry.

What we learned

  • Prompt Engineering is Pedagogy: Crafting a Socratic tutor isn't just about the prompt; it's about defining turn limits, scoring mechanisms, and completion tokens.
  • Visual Trust is User Retention: In EdTech, showing the "how" (thinking steps) is just as important as the outcome (the score).
  • Behavioral Data > Raw Scores: A high score means nothing if the behavioral data (time spent/pattern) suggests a student is on the verge of burnout.

What's next for Eduvoka

  • Persistent Knowledge Graphs: Moving the Concept Map from a session-only visual to a persistent tracker that shows a student's growth over months of study.
  • Study Group Synergy: Implementing multi-student "Socratic Seminars" where Gemini acts as a moderator for group discussions.
  • RAG Integration: Syncing with official UTBK curricula via Retrieval-Augmented Generation to ensure our tutor is always up-to-date with the latest examination standards.

Built With

  • framer-motion
  • google-gemini-api-(gemini-1.5-flash)
  • lucide-react
  • next.js-16-(app-router)
  • nextauth.js
  • postgresql/mariadb
  • prisma-orm
  • radix-ui
  • react-19
  • recharts
  • tailwind-css-4
  • tanstack-react-query
  • typescript
  • zod
Share this project:

Updates