After more than a decade of building crypto platforms and recently leading the Blockchain Council of the Philippines, Luis Buenaventura stumbled across a set of World Bank statistics that derailed his usual routine. The research showed that the average 15-year-old Filipino student possesses the math and science capabilities of a 10-year-old in Singapore, which means local public school children are up to five full school years behind their regional peers.
Rather than waiting for infrastructure or cloud services to bridge the gap, Luis decided to build an AI tutor that works directly on entry-level mobile hardware. The result is Hiraia, an open-source AI science tutor that runs completely offline on $100 Android smartphones, delivering lessons in Tagalog, Bisaya, and English without requiring an internet connection.
“I started working on Hiraia because of this really heartbreaking World Bank research,” Luis said. “We don’t have enough classrooms and we don’t have enough teachers. But it’s 2026, and AI has to be able to do something about this. My hope is that I can roll this out to public schools across the country as a supplemental teaching tool — not to replace teachers, but to reinforce classroom learning with some personal tutoring at home.”
But how did Luis create this AI Science Tutor?
Engineering an AI for $100 Hardware
Most generative AI platforms rely on distant cloud servers, which makes them impractical in a country where over 60 percent of households lack fixed broadband internet. Hiraia solves this by packing the entire intelligence model onto the phone’s internal storage.
- Built on Sea AI Lab’s Sailor 2 model and a continued-pretraining fork of Qwen 3.5-2B, the tutor speaks Tagalog and Bisaya fluently alongside English.
- The app integrates a vector-searched index containing over 40,000 science facts and 30,000 illustrations mapped directly to the Department of Education’s MATATAG curriculum.
- All visuals and facts are pre-generated and verified locally to prevent AI hallucinations. “Every response is based on an established fact in its data bank and every illustration it offers up is pre-generated for child safety reasons,” Luis explained.
Bypassing Mobile Data Barrier
Downloading Hiraia’s full installation package (around 4GB) presents its own hurdle for families without home internet. To solve this, the app incorporates Pears, a peer-to-peer (P2P) file-sharing protocol.
In a classroom setting, only a single phone needs to download the model over a municipal Wi-Fi hotspot. Other students can then transfer the files directly from that device over a local peer-to-peer network without using cellular data. Phones unable to run the background generative AI model can still access Hiraia’s core library of flashcards and quizzes offline.
Funding and Deployment
Luis developed Hiraia backed by a ₱1 million research grant from the Tether Foundation to complete the model’s continued pre-training and fine-tuning.
Currently in early alpha (v0.3.1), the app requires no account registration, displays no ads, and collects zero personal data. Luis is now seeking academic partners to bring the software into real classrooms.
“An on-device inference model is the only viable solution when two-thirds of all households don’t have a fixed internet connection,” Luis noted. “My next big milestone is to partner with an academic institution to safely do pilot trials.”
This article is published on BitPinas: Tech Veteran Built an Offline AI Tutor That Runs on Budget Smartphones
What else is happening in Crypto Philippines and beyond?
