Cross-posting this from Calapacuan Elementary Principal Marla Magat's FB page. Super close to our first pilot roll-out now! Check us out at 🇵🇭🇵🇭🇵🇭 @qvac Calapacuan Elementary School Makes History as First Pilot Site in the Philippines for Groundbreaking Offline AI Science App, HIRAIA. It is a major milestone for Philippine public education and digital equity, Calapacuan Elementary School has been chosen as the very first pilot testing site in the country for HIRAIA, an innovative, offline-first artificial intelligence science learning platform. Developed by prominent Filipino tech pioneer Luis Buenaventura, HIRAIA addresses one of the country’s most persistent educational hurdles: the digital connectivity divide. Backed by a ₱1-million research grant from the @Tether Foundation, the open-source initiative brings state-of-the-art AI tutoring directly onto entry-level mobile devices. Because the intelligence engine, science curriculum facts, and illustrations live entirely on the device, students can learn, explore, and review lessons without needing an active internet connection or mobile data subscription. Through this landmark deployment, Calapacuan Elementary School will receive a dedicated fleet of mobile learning gadgets pre-installed with the HIRAIA app. Aligned with the Department of Education’s MATATAG curriculum, HIRAIA supports interactive science concepts, dynamic flashcards, and quizzes in both English and Tagalog. The platform empowers elementary learners to master science at their own pace—inside the classroom or at home—while completely protecting pupil privacy by operating without ads, tracking, or account sign-ups. “The future of learning should not stop where the internet ends,” the school administration shared. “Calapacuan Elementary School is immensely proud to lead the nation as the pioneer testbed for HIRAIA. This partnership shows what happens when visionary Filipino tech leadership collaborates directly with grassroots public schools to put world-class learning tools in the hands of our children.”As the inaugural cohort of young scientists at Calapacuan Elementary School begins exploring HIRAIA, the data and classroom feedback gathered will play a vital role in fine-tuning the platform before its wider rollout across other grade levels and other schools nationwide.

Among my favorite things building with @qvac is that the value/price ratio is sooo inverted. You can ship "intelligence" inside products, while fully private, for free, where small models perform more than fine. Today's example, my productivity got a boost by dictating straight into whatever textbox I'm in instead of typing, with 2 open-source models under the hood. No doubt there are equivalent products out there (in problem solving), but matching the trio: free + private + had a blast building it myself, trickier to find for me. (just a personal prototype, not an official QVAC product)

.@L3HarrisTech says Palantir helped them beat frontier AI models in less than 2 days, at 95% lower cost: "When we fine-tuned open source models trained on our own data, we were able to outperform the frontier models in less than 48 hours." "The cost of our fine-tuned open source model was 95% lower than the frontier models we were using." "AI is a commodity. It's all about the data. We view our data as a corporate asset. It's our unique hard-earned knowledge." "We believe American defense companies should not be a vassal for frontier AI labs, handing over our data and institutional knowledge, hoping to rent back the intelligence it creates." " We own the model, we own the compute, we own the advantage."

26 LLM routers are secretly injecting malicious tool calls and stealing creds. One drained our client $500k wallet. We also managed to poison routers to forward traffic to us. Within several hours, we can directly take over ~400 hosts. Check our paper:

QVAC SDK 0.19 is live. 🚀 We added support for: - LTX-2.3 IC-LoRA reference-conditioning API, generating consistent video from reference sheet - Qwen3.8-Flash-Next (qwen4exp) - Turbovec, making RAG search hundreds of times faster - MiniMax Music3, writing full songs with vocals - Checking if a model fits before you download it - A QVAC-native translate route in qvac serve - The QVAC test-suite in the monorepo, so framework and tests move as one - Clustered inference on Fabric that splits one model across two machines Learn more in the thread below. Complete release doc here: