Event
LA Hacks
Team
Gavin Huang
Jacob Scheff
Yirui Song
Knova is a fully on-device AI tutor for elementary students built at LA Hacks 2026. Created by Jacob Scheff, Yirui Song and Gavin Huang, Knova won the ZETIC Company Challenge by tackling one of education's most fundamental access problems: 1.3 billion students globally lack reliable internet, and for most of them, learning stops the moment connectivity does.
What it solves
Education today is built on constant connectivity. Cloud-based AI tutors require stable internet, expensive API calls, and infrastructure that billions of students simply do not have. Knova removes every one of those dependencies. No Wi-Fi. No data costs. No interruptions. A safe, personalized, fully self-contained AI tutor that works in airplane mode, in rural areas, and in every pocket it reaches.
What it does
Knova is an offline-first AI tutor that runs entirely on the device, no internet required. Students can upload photos of notes, worksheets or raw text, and Knova transforms them into personalized lessons in real time. A gamified Learning Constellation UI turns every mastered concept into a star, building a visual knowledge map that keeps young students motivated. An on-device LLM personalizes the learning path based on each student's progress over time using Bayesian Knowledge Tracing.
How it works
Knova uses ZETIC to run the Gemma model entirely on-device, enabling fully offline AI inference with low latency and no cloud dependency. Bayesian Knowledge Tracing models each student's understanding over time and dynamically adapts content based on prior interactions. On-device content filtering ensures safe outputs. When connectivity is available, ElevenLabs provides natural voice and Gemini handles more complex queries, with seamless switching between edge and cloud.