A quiet knowledge infrastructure for underserved Filipino learners — offline-first, dignity-first, sovereignty-first.
"You are not building AI tools. You are building scientific dignity."
Understanding what this project truly is before writing a single line of code or buying a single component.
"Science should widen humility, not inflate competition."— The core principle from which everything else flows
Just as a UPS stabilizes power through brownouts, this system stabilizes cognitive access through infrastructure gaps. Survival stress compresses perspective. Science expands it — placing hardship inside cosmic scale without diminishing dignity.
Not a search engine. Not a chatbot. A steady, patient presence — like a lighthouse that doesn't chase ships, but makes wisdom quietly available to those who seek it. The website, the kit, and the content all serve one function: being findable when needed.
Internet = enhancement, never dependency. This system operates in areas where Manila infrastructure does not reach. It is designed around this reality — not as a limitation, but as a core feature. Offline-first is not a compromise. It is the product.
Filipino learners in provinces should not feel that knowledge is a foreign luxury. This kit contextualizes science inside Filipino culture — humility + scale awareness = grounded strength. Not chasing dominance. Building belonging.
A layered, resilient architecture built on the principle: local-first, cloud-never-required.
Everything fits in a small carry case. A teacher can set it up in under 5 minutes. A UPS in the classroom for knowledge.
Silent form-factor. Fits in a backpack. Powers the entire AI system.
Sustains at least 20 min runtime on outage. Same thinking as data center — stability is dignity.
Creates a local hotspot. Students connect via browser. No internet required — just LAN.
Optional for off-grid barangays. Paired with a battery controller for stable DC input.
Optional classroom monitor for teacher dashboard and group Q&A mode.
One for USB OTA updates, one for backup. Essential for zero-internet areas.
| Component | Spec | Est. Cost (PHP) | Priority |
|---|---|---|---|
| Mini PC | Intel N100, 16GB RAM, 512GB SSD | ₱8,000 – 14,000 | REQUIRED |
| RAM Upgrade | To 32GB DDR5 (for larger models) | ₱3,000 – 5,000 | UPGRADE |
| NVMe SSD 1TB | For full model + curriculum storage | ₱2,500 – 4,000 | REQUIRED |
| UPS 650VA | APC BX650 or equivalent | ₱3,500 – 5,000 | REQUIRED |
| WiFi AP | TP-Link EAP or Ruijie RG-AP | ₱2,000 – 4,500 | REQUIRED |
| USB Drives (2x) | 64GB USB 3.0 — update + backup | ₱600 – 1,000 | REQUIRED |
| Carry Case | Foam-padded, lockable | ₱1,500 – 3,000 | REQUIRED |
| Solar Panel Kit | 50W panel + PWM controller | ₱4,000 – 7,000 | OPTIONAL |
| TOTAL (Core) | Without solar, without monitor | ₱21,100 – 36,500 | FEASIBLE |
* Prices based on Lazada/Shopee PH and local IT suppliers as of early 2025. Kit can be assembled at ₱25,000 target.
Three clearly defined operating states. The experience is identical for students in all three — the system handles the rest silently.
The operational heartbeat of the system. Silent, patient, and fail-safe. A process that never interrupts learning.
Via browser on phone/laptop connected to local WiFi. Asks in Filipino or English. No login required.
0ms network latency — all localInput passes through age-appropriate filter and guardrails. Bias-awareness module flags sensitive framing. Calm, non-threatening response mode enforced.
Child-safe by designQuery is embedded and matched against local FAISS vector database containing Philippine curriculum, science modules, and AI literacy content.
Context-grounded answers7B quantized model generates response using retrieved context. Temperature tuned low for educational accuracy. Max 400 tokens for clarity.
Fully local · No cloud callStudent sees answer in browser. Calm, plain language. No flashy animation. Response logged locally, encrypted. Purged after 30 days.
Simple · Clear · DignifiedBackground process activates. Does not wake users. Does not interrupt any active session. Completely transparent to the classroom.
No UI notification everPings lightweight health-check endpoint. If no response in 3 seconds — skip entirely. Resume offline mode with zero change.
Fail-silent · No alarmIf internet found, checks available bandwidth. Below threshold (e.g. 512kbps)? Skip this cycle. Reschedule to next 6-hour window.
Smart schedulingDevice authenticates with API key. Downloads delta patch package (not full redownload). Verifies SHA-256 checksum before applying. Rolls back if verification fails.
Verified · Tamper-proofModel delta applied. FAISS vector index rebuilt from new curriculum packs. Total downtime: under 2 minutes. System restores automatically. Logs sync result locally.
Seamless update · Auto-restoreAfter every brownout or power failure — the system must restore itself without teacher intervention.
Three role-based views: Student, Teacher, and Admin. Accessible via any browser on the local WiFi network.
Additions and refinements to strengthen the original framework — preserving its soul while sharpening its structure.
The philosophical grounding is rare and powerful. Most ed-tech projects lead with hardware specs. This one leads with dignity. The UPS analogy is perfect — it teaches the product's purpose in one sentence. The three-mode network design is technically sound and production-grade. The fail-silent principle is correct and caring. The data sovereignty approach (no personal data, auto-purge) will be essential for DepEd and NGO trust. The USB manual update path is an often-missed but critical inclusion for zero-connectivity areas.
Filipino and English are assumed, but the Philippines has 8 major regional languages. Visayas and Mindanao classrooms speak Cebuano and Hiligaynon. Adding even basic language detection and response in these languages would dramatically expand reach and emotional belonging. The model can be prompted to respond in dialect when detected.
PRIORITY: HIGHTeachers should be able to set a "lesson context" for the day (e.g., "Chapter 4: Solar System") so that student questions are automatically scoped and contextualized. This prevents off-topic rabbit holes and ties AI answers back to curriculum. The LLM receives the lesson context as a system prompt for that session window.
PRIORITY: HIGHTrack which topics students ask about most frequently — in aggregate, never individually. This data helps curriculum designers understand what Filipino students are actually curious about vs. what is taught. The heatmap should be visible to teachers and, when connected, to a central research dashboard. This is scientifically valuable data, not surveillance.
PRIORITY: MEDIUMThe "<60W continuous" target is good but needs to account for peak loads. An Intel N100 Mini PC at full LLM inference load can hit 30–40W. Add WiFi AP at 8–12W, UPS overhead, and display if present — total can reach 70–80W peak. Recommend specifying a 1200VA UPS (not 650VA) if solar input is not available, and defining a thermal throttle policy for sustained inference sessions in warm classroom environments (common in Philippine provinces without aircon).
PRIORITY: HIGH — Infrastructure RiskThe API key / token authentication described in Mode 2 is correct but needs expansion. Recommend: (1) Device fingerprint at provisioning time, (2) Rotating token system (30-day expiry), (3) Update package encryption with device-specific public key. This ensures that even if a USB update drive is stolen, the package cannot be replayed on an unauthorized device. Critical for DepEd trust.
PRIORITY: HIGH — Trust RequirementDefine a reproducible deployment standard: a single USB stick that bootstraps the entire system from a bare Mini PC in under 20 minutes. Include a setup checklist printed inside the carry case lid. This enables NGOs, LGUs, and barangay volunteers to deploy kits without technical expertise. The democratization of the deployment, not just the AI, is what creates scale.
PRIORITY: HIGH — ScalabilityThe system needs a name and a character — not a corporate AI, but a humble Filipino guide. "Liwanag" (light/clarity in Filipino) fits perfectly with the lighthouse philosophy. Give it a consistent voice: patient, curious, never condescending. "Liwanag" references Philippine heroes, local geography, and cultural examples in its answers. This transforms a tool into a companion — and companions get used.
PRIORITY: MEDIUM — Adoption DriverLong-term vision: when multiple Liwanag kits are deployed across provinces, enable a privacy-preserving federated learning update. Each kit contributes anonymized interaction patterns to improve the shared model — without any personal data leaving the device. This creates a model that becomes more Filipino over time, trained by Filipino students, for Filipino students. This is the true network effect of dignity at scale.
PRIORITY: LONG-TERM VISION"Growth is not found in the noise of the Awakening,— AI Awareness Inspired · The heart of everything this system does
but in the steady, patient rhythm of Awareness-Inspired utility."