AUGUST 25, 2025
better-ed joined the Google Gemma 3n Impact Challenge to reimagine assessments with voice, offline AI, and equal access to learning for all students.

Jovi Maniago
Head of Marketing at Better-ed
In July 2025, Predictive Systems Inc. (PSI) proudly submittedbetter-ed, our voice-enabled AI assessment platform, to theGoogle – Gemma 3n Impact Challenge on Kaggle. The competition gathered innovators worldwide to explore how Google’s Gemma models could be applied to real-world problems. For us, it was the perfect opportunity to showcase how AI can transform assessments into authentic, conversational experiences that make learning truly heard.
The hackathon invited participants to harnessGemma, Google’s family of lightweight open models, to create solutions with meaningful social impact. Unlike conventional Kaggle contests focused purely on accuracy, this challenge emphasizedcreativity, usability, and inclusivity.
Our submission video captured the essence ofbetter-ed:
For readers new to the concept,Gemmais Google’s family ofsmall, efficient, and open-source AI models. They are designed to be:
For education, this means that advanced AI doesn’t have to stay locked behind expensive servers or fast internet connections. Gemma opens the door tooffline, local AI solutions—something we believe is essential for making learning equal and accessible everywhere.
At better-ed, we believe education should nurture not just what students know, but how they express what they’ve learned. Our mission is rooted in three commitments:
We entered the competition with clear goals:
Here, Gemma ties directly into our mission. Many schools, especially in rural areas, face challenges with connectivity. By leveragingGemma’s lightweight, offline-friendly capabilities, better-ed can provide:

This innovation ensures no student is left behind—whether they’re in urban centers or far-flung provinces.
Our submission video also featured an interview with Allan Tan, PSI’s founder and CEO, where he shared how Gemma 3npowers better-ed’s approach to voice-based learning.
The team needed a small, efficient model that could run offline on a phone and support voice. Among the options evaluated, Gemma 3n delivered the best performance, especially when paired with Unsloth for faster response times. This made it possible to provide real-time experiences in classrooms without relying on heavy infrastructure.
Students answer by speaking, and Gemma handles the transcription and response generation. Since Gemma doesn’t natively perform text-to-speech, the team prerecorded a set of teacher-style responses. The model then selects the most appropriate one, making the interaction feel conversational and natural.
To avoid shallow answers, the team tested the system against simulated student personalities—from grade A learners to average students, and even those who didn’t study or were just playing around. This ensures better-ed can recognize a wide range of real-life classroom scenarios and give teachers authentic insights into student learning.
This behind-the-scenes view highlights how Allan and the team deliberately designed better-ed not just to process responses, but to reflect the diversity and complexity of real classrooms.
The Gemma 3n Impact Challenge reaffirmed three things for us:
In the video, you’ll see how voice transforms assessment, Allan’s reflections on our mission, and a preview of Gemma-powered offline capabilities.
The challenge was a milestone, but our journey continues. We are scaling pilots, refining offline features, and working closely with educators to shape the future of voice-based learning. Every step brings us closer to a world where education isn’t just tested, but truly heard.
Through theGoogle Gemma 3n Impact Challenge, we showed that better-ed is more than a platform—it’s a movement to reimagine assessments with AI that listens.
At PSI, we remain committed to the principle:Your Data. Your AI. Your Way.
Curious about how better-ed can transform your classroom or institution?Get in touch with PSI today to explore tailored AI solutions that make learning heard.