Inspiration
Each year, over 1.8 million students in Nigeria sit for the UTME (Unified Tertiary Matriculation Examination), yet access to quality preparation remains deeply unequal. Traditional tutoring centers charge up to ₦150,000 ($100)—unaffordable for most families—while existing mobile apps largely provide static question banks with no personalization, feedback, or conceptual guidance.
We were inspired to build Quizant after observing that students don’t fail due to lack of effort, but due to lack of guided understanding. We asked a simple question:
What if every student could have a personal AI tutor—one that understands their weaknesses, explains patiently, and adapts in real time?
What it does
Quizant is an adaptive AI-powered learning platform designed for high-stakes exam preparation:
- Savant AI Tutor: A multi-turn conversational tutor powered by Gemini 3 reasoning, designed to behave like a Socratic teacher—guiding students step by step instead of giving answers outright.
- Flashcard Generator: Students upload PDFs or photograph handwritten notes, which Gemini’s Vision capabilities convert into structured, high-quality digital flashcards.
- Quiz Generator: Generates unlimited, curriculum-aligned practice questions with detailed explanations and structured feedback.
- Concept Visualizer: Uses Gemini’s reasoning to convert abstract topics into interactive Mermaid.js diagrams, supporting visual learners.
How we built it
Quizant was built with React + TypeScript on the frontend and Tailwind CSS for a clean, accessible UI. The backend uses Supabase for authentication, storage, and real-time data syncing.
The core intelligence layer is built entirely on Gemini 3.5 Flash Preview, used in four critical ways:
- Streaming API: Enables token-by-token responses in the Savant AI Tutor, creating a natural, conversational learning experience.
- Vision API: Powers document ingestion by extracting structured knowledge from handwritten and printed notes.
- Structured JSON Output: Ensures quizzes, flashcards, and feedback follow strict schemas for reliability and consistency.
- Reasoning Capabilities: Generates Mermaid.js syntax for concept diagrams and structured explanations.
We implemented LaTeX rendering to accurately display mathematical and scientific expressions, such as:
$$ F = ma $$
and advanced integrals like:
$$ \int_0^\infty e^{-x^2} dx = \frac{\sqrt{\pi}}{2} $$
Challenges we ran into
One of the hardest challenges was combining real-time AI streaming with LaTeX rendering. Partial LaTeX tokens (e.g. \frac{) arriving mid-stream could break rendering and crash the UI.
We solved this by implementing a custom Markdown and LaTeX buffering system that waits for syntactically complete tokens before rendering, ensuring smooth real-time explanations without visual glitches.
Accomplishments that we’re proud of
- Built a voice-enabled tutoring interface, allowing students to speak directly to the AI tutor.
- Reduced document-to-flashcard processing time to under 5 seconds using Gemini Vision.
- Designed a premium, distraction-free UI that students described as feeling “magical” rather than overwhelming.
- Successfully used Gemini’s structured output to eliminate formatting errors in quizzes and feedback.
What we learned
We learned that prompt engineering is systems engineering. Achieving high educational value required carefully designed system prompts that enforced a Socratic tutoring style rather than answer-dumping.
We also learned how powerful Gemini’s structured JSON outputs are for building dependable, production-grade educational tools that require consistency and trust.
What’s next for Quizant
Our next goal is scale and accessibility. We plan to:
- Partner with local JAMB centers to deploy Quizant in underserved regions.
- Add offline-first support and SMS-based quizzing for students with limited internet access.
- Use Gemini’s efficient caching to reduce bandwidth usage and data costs.
Our long-term vision is simple: make high-quality, personalized education available to every student—regardless of income, location, or infrastructure.
Built With
- api
- css
- framer
- gemini
- mermaid.js
- motion
- react
- recharts
- supabase
- tailwind
- typescript
- vercel
Log in or sign up for Devpost to join the conversation.