BORG
Hybrid Digital Sovereignty Platform
University of Akureyri's secure hybrid AI platform — combining commercial AI with on-premise GPU infrastructure for academic data sovereignty.
The Challenge
Universities need cutting-edge AI capabilities for research, teaching, and administration, but academic data — student records, research materials, institutional communications — demands strict privacy guarantees. Commercial AI services offer capability and scale, but sending sensitive institutional data to external APIs creates compliance and sovereignty risks. Most institutions either ban AI outright or adopt it without governance — both of which fail students and staff. The University of Akureyri needed a platform that resolved this tension: full AI capability with full data sovereignty.
The Approach
BORG implements a hybrid architecture that resolves this tension through intelligent routing. Commercial AI providers (OpenAI, Anthropic, Google) handle non-sensitive workloads, while an on-premise GPU workstation — 'Hercules', with RTX 5090 + RTX 4090 totalling 56GB VRAM — runs private language models via Ollama for data that must never leave campus. The platform is built on Next.js 16 with React 19 and PostgreSQL via Prisma 7, deployed across multiple virtual machines on the university's internal network (VPN-required). Authentication uses Authentik SSO with university identities and role-based access control. A modular Cubes extension framework allows domain-specific tools to be developed and deployed independently without modifying the core platform. Each Cube gets authentication, API routes, and its own configuration. The VoltAgent 2.x abstraction layer provides a unified interface for all AI providers with intelligent model selection: fast models for routine tasks, reasoning models for complex analysis, and local models for private inference.
Outcomes
BORG is a production AI platform built for the University of Akureyri. It resolves a tension that most institutions avoid: how to give staff and students full AI capability without sending sensitive academic data to external providers. The answer is hybrid sovereignty — commercial AI for general workloads, on-premise GPU for everything that must stay on campus.
The platform has been in production since early 2024 and serves the university's research, teaching, and administrative needs across seven integrated tools — from an AI chatbot with citation-backed RAG to a syllabus evaluator that checks compliance with Bologna Process standards.
Architecture
The platform is built on Next.js 16 with React 19 and PostgreSQL via Prisma 7, deployed across eight virtual machines on the university's internal network. Access requires VPN. Authentication uses Authentik SSO with university identities and role-based access control — staff, students, and administrators see different capabilities based on their institutional role.
- Commercial AI — OpenAI, Anthropic, and Google handle non-sensitive workloads where capability and speed matter most. The VoltAgent 2.x abstraction layer provides a unified interface with intelligent model selection: fast models for routine tasks, reasoning models for complex analysis.
- On-Premise GPU — 'Hercules', a dual-GPU workstation with RTX 5090 and RTX 4090 totalling 56GB VRAM, runs private language models via Ollama. Student records, research data, and institutional communications never leave campus.
- Infrastructure — Full production, staging, and development environments with nginx reverse proxy, internal DNS, separate database servers, and automated CI/CD via GitHub Actions.
Platform Tools
Seven tools serve distinct university needs. Each is built as an independent Cube — a modular extension that gets its own authentication, API routes, and configuration without modifying the core platform.
- AI Chatbot — Multi-provider RAG with inline citations. Supports document upload, conversation history, and intelligent routing between commercial and local models based on data sensitivity.
- Speech-to-Text — Icelandic and English transcription with custom pipelines. Icelandic language support requires dedicated infrastructure that commercial providers do not yet offer at production quality.
- Syllabus Evaluator — Evaluates course syllabi against Bologna Process learning outcome standards. Domain context creates more value than raw model capability — this tool demonstrates why university-specific AI matters.
- Media Transformer, Podcast Studio, Admin Dashboard — Audio/video conversion, browser-based podcast recording, and administrative oversight of platform usage and model costs.
Governance by Design
BORG is not just a technical platform — it is the university's AI governance infrastructure. The hybrid architecture enforces data sovereignty at the routing level: sensitive data physically cannot leave campus because the routing logic prevents it. This is governance through architecture, not policy documents.
The platform embodies the thesis argued in Beyond Fragmentation: that AI governance fails not from a lack of regulation but through structural mechanisms — Responsibility Fog and Cognitive Debt — that erode accountability and human judgment. BORG addresses both by making sovereignty architectural and keeping humans in control of what the AI can access.
Technology Stack
Lessons Learned
- Hybrid sovereignty — combining commercial and local AI — resolves the false choice between capability and privacy. You don't have to pick one.
- A modular extension framework prevents the platform from becoming a monolith while enabling rapid feature development.
- Icelandic language support requires dedicated infrastructure (custom speech-to-text pipelines) that cannot rely solely on commercial providers.
- University-specific tools like syllabus evaluation against Bologna Process standards demonstrate that domain context creates more value than raw model capability.
- An AI provider abstraction layer future-proofs against lock-in — when costs drop dramatically year over year, you need to switch models without rewriting your platform.
Explore the Project
Related Projects
View all projects →
Smarason.is
A privacy-first portfolio built on Next.js 15, Tailwind v4, and Sanity CMS. Bilingual, accessible, zero tracking — the architecture reflects the philosophy.
Beyond Fragmentation
MA thesis on AI governance. Introduces Responsibility Fog, Cognitive Debt, and the VALOR Framework.
Sumarhus Alpha
The infrastructure behind a personal AI system: 34 skills, 16 hooks, persistent memory, multi-agent collaboration, and a security model forged through documented failure. Evolving daily since 2022.