Synapse

💡 Inspiration

In Latin America and around the world, most AI educational tools act as simple answer generators. Students copy-paste their homework, the AI gives them the direct answer, and no actual learning occurs. Furthermore, millions of students face cellular data limitations that prevent them from accessing heavy web apps, although they have unlimited access to messaging platforms like WhatsApp.

We built Synapse to bridge this gap: an AI-powered virtual tutor driven by Google Gemini 2.0 Flash that applies the Socratic Method (guiding students through questions rather than giving direct answers) and reaches students wherever they are—including WhatsApp—while providing teachers with real-time analytics on student knowledge gaps.

🧠 What it does

Synapse transforms passive homework copying into active, guided learning through 5 core pillars:

Socratic AI Tutor: Synapse never gives away direct solutions. When a student asks a question or gets stuck on an equation like: 2x + 5 = 15 Synapse breaks down the problem into logical steps and asks guiding questions , leading the student to discover the answer themselves.
Real Multimodal Feedback:
- Notebook Photo Analysis: A student takes a photo of their handwritten notebook page. Gemini 2.0 Flash analyzes the image, identifies the exact line where a mistake occurred, and gently points out the specific step to review.
- Voice Notes: Frustrated students can send voice notes. Synapse listens, assesses their comprehension level, and responds using simple daily analogies and guiding questions.
Adaptive Quizzes: Generates dynamic 3-to-5 question micro-quizzes tailored specifically to each student's current weak spots.
Teacher Knowledge-Gap Dashboard: Aggregates student interactions to provide educators with actionable real-time analytics (e.g., "65% of your class struggled with algebraic sign inversion this week").
Universal Accessibility via WhatsApp: Delivers Gemini 2.0 Flash directly into WhatsApp, eliminating data barriers for students across LATAM.

🛠️ How we built it

Core AI: Powered by Google Gemini 2.0 Flash API (gemini-2.0-flash) for low-latency multimodal reasoning, structured JSON outputs, and fast conversational interactions.
Backend: Built with Node.js and Express.js, modularized into specialized AI agents (tutorSocratico.js, quizGenerator.js, audioProcessor.js).
Database & Auth: Firebase Firestore for real-time conversation history, student profiles, and knowledge gap aggregations.
Channels & Webhooks: Integrated via WhatsApp Business Webhooks alongside a clean Web frontend for student and teacher dashboards.

🚨 Challenges we faced

Enforcing the Socratic Constraint: Prompting LLMs to refrain from disclosing final answers required rigorous prompt engineering and system guardrails so Gemini consistently responds with guiding micro-questions.
Parsing Handwritten Notebook Images: Ensuring accurate error identification in varied handwriting and low-quality photo uploads.
Structured Analytics Extraction: Extracting structured action tags (e.g., [VACIO_DETECTADO|concepto:...]) smoothly from real-time streaming conversations without degrading the natural tone.

🎓 What we learned

Pedagogy-first AI design creates significantly higher cognitive engagement than simple QA bots.
Gemini 2.0 Flash's multimodal speed makes real-time image error diagnosis and audio interaction feel instantaneous and natural.
Designing with regional accessibility constraints (e.g., WhatsApp integration) unlocks AI benefits for underserved communities.

🚀 What's next for Synapse