Inspiration

Growing up in a small Haryana town, I witnessed my cousins walk five kilometers every day to school, only to sit in classes with over eighty students and a teacher who was overworked. The statistics hit harder in 2025: 20% of rural children drop out before Class 10, 44% of rural children still lack proper teachers, and the majority of EdTech apps are designed for urban English-speaking students with 4G. I couldn't stop wondering why India's 1.2 billion mobile phones couldn't be used as village teachers. That hope and rage turned into LUCID.

What it does

LUCID is a voice-first, fully offline-capable AI tutor designed exclusively for rural Indian students (Class 5–10). Speaks & understands 22+ Indian languages/dialects (via Bhashini) Works offline – caches 50+ lessons, syncs when network returns Pauses DIKSHA videos on “Doubt!” and answers in <2 seconds with voice + AR diagrams Adapts difficulty in real-time using scikit-learn + emotion detection (MediaPipe) Sends dropout-risk SMS alerts to parents/teachers (Twilio) Costs only ₹300/month – uses free DIKSHA/SWAYAM content Gives teachers beautiful Pandas dashboards to track entire classes

How we built it

*One developer, four intense weeks. * Frontend React Native + Expo + Redux → offline-first, voice-first UI Expo AV + Gifted Chat → video + chat experience TensorFlow.js → on-device emotion & quiz logic (no internet needed) Backend & AI FastAPI + Hugging Face Transformers (Indic models) LangChain + Neo4j Knowledge Graph → KG-RAG to keep answers NCERT-accurate Bhashini API → real-time translation for 22+ languages DIKSHA & SWAYAM APIs → 100% free government content Kubernetes on AWS Mumbai → ready to scale

Challenges we ran into

Training multilingual models with almost zero rural dialect data → solved with synthetic voice data + Bhashini fine-tuning 2G networks dropping every 10 seconds → built aggressive caching + resumable downloads Students speaking slang/mixed languages → added fallback fuzzy matching + teacher correction loop Parents afraid of “robot teacher” → added warm Hindi voice + “Your child asked about photosynthesis today” SMS feature

What we learned

Voice > text for rural India (retention jumped 2.5× when we removed typing) Emotion detection actually works – when the AI says “Aram se, samjhaata hoon” after detecting frustration, kids smile and continue Government APIs (DIKSHA, Bhashini) are gold if you’re patient with documentation Never underestimate the power of a ₹300 price point in villages

What's next for LUCID

Add AR science experiments (virtual dissection, physics sandbox) – Q1 2026 Peer-to-peer study groups inside the app Partner with state governments for free deployment in 100,000 schools Expand to Class 11–12 + competitive exams in regional languages

*LUCID is more than just an app. Finally designed for the India that was falling behind, it is the teacher who never sleeps, never passes judgment, and speaks your mother tongue. *

Built With

Share this project:

Updates