Dura — The Knowledge Granary

Research documentation and collaboration platform for privacy-preserving technology

A living knowledge base making complex science and technology accessible to everyone, inspired by the spirit of Dr. Alex Magaisa's Big Saturday Read.

Live Site: https://dura.disruptiveiot.org

---

🎯 Purpose

Dura is part of the Disruptive IoT, exploring The Economy of Things (EoT), the concept of IoT devices becoming economic participants, not just data collectors.

"IoT connected devices. EoT disrupts how they earn."

When devices become economic actors—with wallets, the ability to transact, and autonomous decision-making, everything changes. Dura scientifically uses AI agents to document the convergence of:

• Blockchain — Economic Agency (devices with wallets)
• AI — Cognitive Agency (devices that think)
• Edge — Operational Agency (devices that process locally)
• IoT — Physical Reality (the hardware layer)

Dura transforms complex blockchain, cryptography, and IoT research into accessible, interactive documentation. We bridge the gap between dense academic papers and practical understanding through:

• Simplified Explanations: Complex concepts made accessible to everyone
• Research Papers: Curated collection with starring and library features
• Progress Tracking: Track your learning journey across content
• Collaborative Research: Contribute and collaborate on research documentation

The Problem We Solve

Blockchain research documentation typically fails in three ways:

1. Inaccessible Theory: Dense whitepapers with no practical examples
2. Disconnected Practice: Code implementations divorced from conceptual foundations
3. Static Content: PDFs and wikis that can't keep pace with rapid research iteration

We bridge theory and practice through interactive concept cards that show:

• Simple analogies (e.g., "Where's Waldo" for ZKPs)
• General cryptography principles
• Project-specific implementations (actual circuits, protocols, architectures)

The Two Sides of Dura

Side	Audience	Content
Public ("Citizen Scientist")	General public, curious minds	Simplified explainers, analogies, practical implications
Research Layer	Researchers, collaborators	Primary papers, technical specs, protocols, peer notes

---

🏗️ Projects Documented

EdgeChain

Privacy-preserving agricultural AI for smallholder farmers

The Problem: 96% of African farmers see digital agriculture as beneficial, yet adoption remains minimal due to justified fears of data exploitation. The "Nyakupfuya" (Shona: trapped farmer) faces surveillance from governments, NGOs, and agribusinesses.

The Solution: EdgeChain uses Zero-Knowledge Proofs and federated learning to transform data from a weapon of punishment into infrastructure for cooperative empowerment.

Concept	Application
ZKP	ProveYieldAboveThreshold circuits — farmer proves yield > quota without revealing actual harvest
KACHINA	Compact smart contracts
Disclosure Regime	Selective compliance proofs

Constraints: 2G network compatibility (SMS-first), 70% of target users live in poverty, deep distrust of data collection. Currently research-only (legal constraints until mid-2027).

Repo: github.com/solkem/edgechain

---

Msingi

Anonymous IoT device identity and autonomous wallets

Core Thesis: Anonymity requires device-held keys, and ZK-enforced spending policies make device wallets safe. If a human wallet pays for a device's transaction, the payment links device to human identity, destroying all privacy guarantees.

Three Pillars:

1. BRACE — Blind Registration (device proves it's registered without revealing which device)
2. IAR — Incentivized Anonymous Relay
3. ACR — Anonymous Contribution Rewards

Integration: Msingi sensors submit anonymous attestations consumed by EdgeChain's Dual Merkle Root system to verify IoT data separately from manual farmer reports.

Repo: github.com/solkem/msingi

---

Ndani

Trustless Privacy-Preserving Infrastructure for Machine Economies

The Problem: Existing IoT-blockchain systems (IOTA, Helium) allow devices to earn, but payments flow to human-controlled wallets. Privacy solutions rely on trusted gateways — if compromised, privacy is lost.

The Solution: Ndani (Swahili for "Inside") moves the entire trust boundary inside the farmer's control:

1. Device-Held Wallets — Keys live in the ATECC608A secure element
2. Local Proof Server — Farmer-owned Raspberry Pi 5 generates ZK proofs locally

Hardware Stack (~$163.50):

Component	Function	Cost
ESP32-S3	Sensor Microcontroller	$8
ATECC608A	Identity + Wallet Keys	$2.50
SX1276	LoRa Transceiver	$6
Raspberry Pi 5	Local Proof Server	$80
Sensors/Power/Infra	Environment + Solar	$57

Architecture: [Sensor Node] <-- LoRa --> [Local Pi 5 Proof Server] <-- Internet --> [Midnight Network]

Repo: github.com/solkem/ndani

---

🤖 AI Agents & Nyakupfuya Memory System

Dura uses AI agents to curate and synthesize research papers, making them accessible to Citizen Scientists. The system learns from human feedback through the Nyakupfuya Memory System.

Agent Architecture

PDF Upload → Extract Text → Curator Agent → Synthesizer Agent → Save to Database
                                ↓                    ↓
                          [Learns Patterns]    [Learns Patterns]
                                ↓                    ↓
                          Pending Memories → Admin Review → Validated Memories
                                                               ↓
                                                    Future Processing Uses Memories

Agents

Agent	Purpose
Curator	Evaluates papers for relevance and accessibility. Runs the "Nyakupfuya test" — can this be explained to a livestock keeper?
Synthesizer	Generates summaries, key concepts, analogies, and identifies paper relationships

Nyakupfuya Memory System

Named after the Shona word for a farmer bound by debt, the memory system represents liberation through accessible knowledge. It's a human-in-the-loop learning system where:

1. Agents create observations during paper processing
2. Admins review at /admin/memories
3. Validated memories influence future processing

Memory Types

Type	Purpose	Example
Episodic	Specific events	"Paper X was rejected because it lacks practical applications"
Semantic	General knowledge	"Privacy papers often require explaining ZK proofs first"
Procedural	How-to patterns	"When summarizing for farmers, use livestock analogies"
Reflective	Meta-observations	"Users prefer shorter paragraphs in Nyakupfuya summaries"

Memory Schema

CREATE TABLE memories (
    id TEXT PRIMARY KEY,
    type TEXT NOT NULL,           -- episodic, semantic, procedural, reflective
    content TEXT NOT NULL,        -- The learned pattern
    confidence REAL DEFAULT 0.5,  -- Agent confidence score
    status TEXT DEFAULT 'pending', -- pending → validated/rejected
    source_agent TEXT,            -- curator, synthesizer
    evidence TEXT,                -- JSON: { paperIds: [], observations: [] }
    reviewed_by TEXT,
    reviewed_at TEXT,
    review_notes TEXT
);

Paper Processing Flow

1. Upload PDF at /admin/papers
2. Curator evaluates:
   • Domain relevance (0-1)
   • Accessibility score (0-1)
   • Creates pending memories for notable patterns
3. Synthesizer generates:
   • One-liner summary
   • Paragraph summary
   • Nyakupfuya (rich, multi-paragraph explanation with village/farming analogies)
   • Key Concepts with definitions and analogies
   • Practical Implications
   • Learning Path (prerequisites, next steps, questions)
   • Creates pending memories for concept explanations
4. Paper saved to database
5. View in library at /papers

---

🛠️ Tech Stack

Layer	Technology	Purpose
Framework	Astro 4.x	Hybrid SSR/SSG with partial hydration
Content	MDX	Markdown with embedded React components
Database	SQLite + Drizzle ORM	User progress, starred papers, sessions
Auth	Lucia	Session-based authentication
Search	Pagefind	Static search indexing
Styling	Tailwind CSS	Utility-first styling
Hosting	DigitalOcean (Docker)	Self-hosted with Nginx reverse proxy

---

📁 Project Structure

dura/
├── src/
│   ├── components/          # React + Astro components
│   ├── content/             # MDX content (learn, papers, projects)
│   ├── db/                  # Database schema & connection
│   ├── layouts/             # Page layouts
│   ├── pages/               # Routes (including API)
│   └── auth.ts              # Lucia authentication
├── drizzle/                 # Database migrations
├── public/                  # Static assets
├── Dockerfile               # Production container
└── .github/workflows/       # CI/CD (deploy.yml)

---

🚀 Quick Start

Prerequisites

• Node.js 20+
• npm

Local Development

# Clone the repository
git clone https://github.com/solkem/dura.git
cd dura

# Install dependencies
npm install

# Start dev server (http://localhost:4321)
npm run dev

# Build for production
npm run build

---

🔐 Deployment

Production: Deployed on DigitalOcean Droplet via Docker + Nginx.

Auto-Deploy (GitHub Actions)

Push to main → GitHub Action builds Docker image → Pushes to GHCR → SSHs to server → Pulls and restarts container.

Environment Variables (Production)

• DB_URL=/app/data/dura.db — Persistent SQLite database
• NODE_ENV=production

SSL Configuration

Uses Let's Encrypt certificates with Nginx. See .dura-migration/ssl_sni_case_study.md for SNI configuration details.

---

✨ Features

• User Authentication: Sign up, login, logout with secure sessions
• Progress Tracking: Mark articles as read, track learning progress
• Paper Library: Star and save research papers
• Search: Full-text search across all content (Pagefind)
• Responsive Design: Works on mobile and desktop

---

📊 Architecture Decisions

• Astro: Ships minimal JavaScript, excellent for content-heavy sites
• SQLite: Simple, file-based database perfect for self-hosted deployment
• Lucia: Modern session-based auth (no JWT complexity)
• Docker: Consistent deployments with volume persistence

---

👥 Team

Project Lead: Solomon Kembo

---

📚 Resources

• Live Site: dura.disruptiveiot.org
• GitHub: github.com/solkem/dura

---

Last Updated: January 2026 Version: 0.3.0