SIH 2026 • Team M1racles (PS SIH26042)// 2026• mobile
Vani
Hindi to Santhali AI Teaching Assistant
Hindi to Santhali teaching assistant Flutter mobile app created for Smart India Hackathon (SIH 2026, team M1racles, Problem Statement SIH26042). The online prototype currently runs via IndicTrans2; offline pipeline targeting quantized NLLB-200M + Qwen2.5 INT4 (under 750 MB, 2s latency) is in development.
OPINIONATED ARCHITECTURAL DECISION
Designing for an edge-native on-device offline translation pipeline to serve remote schools in tribal belts without stable internet connectivity.
WHAT I GOT WRONG & LESSONS LEARNED
Initial unquantized models required over 2.5 GB of RAM, immediately failing low-end mobile devices and demonstrating why INT4 quantization constraints were mandatory.
How I built this & what it delivers
Built Flutter mobile client for Hindi to Santhali curriculum teaching for SIH 2026.
Live prototype runs online using IndicTrans2 for dialect-aware translation.
Designing offline edge pipeline with quantized NLLB-200M and Qwen2.5 INT4 targeting <750 MB bundle size and <2s latency.
Developed as part of team M1racles for Problem Statement SIH26042.
// TECH ARSENAL
FlutterDartIndicTrans2NLLB-200MQwen2.5INT4 QuantizationOffline AI
// PERFORMANCE & IMPACT
SIH 2026 (PS SIH26042) • IndicTrans2 Prototype • <750MB Offline Pipeline Planned
Audited for production reliability, memory efficiency, and execution safety.