Handbook for Technology and Organisation

Innovation

Innovation

New technology counts only once benefit, risk and operation are proven

Innovation evaluates new technologies by benefit, risk, and operational viability. Generative AI, decentralised architectures, and quantum computing change requirements for data, security, and governance. Resilient enterprise architectures such as RAG, Data Mesh, and Sovereign AI are the relevant target state.

This module describes the evaluation and implementation of new technologies. It covers technology management processes (Tech Radar) and ethical and regulatory guardrails (Digital Ethics) in an AI-shaped economy.

Three Innovation Guiding Principles

Technology decisions follow three principles:

  • Data-Driven Architecture: Data is a productive asset. Modern architectures (Data Mesh, Medallion) make data available decentrally and centrally auditable.
  • Generative AI and RAG Architecture: The use of AI requires protection of intellectual property. Retrieval-augmented generation (RAG) on controlled infrastructure can connect AI use with clear data flows and lower dependency on public models.
  • Resilience through Decentralisation: Emerging technologies (Web3, edge computing) are useful where they reduce dependency on central platform operators and improve fault tolerance.

Table of Contents

  • Data Architecture: Reliable insights need structured data pipelines. Medallion Architecture and Data Mesh turn raw data into verifiable, scalable foundations for AI.
  • Modern Databases: Modern applications need the right database for each task. Polyglot Persistence combines SQL, NoSQL, Graph and Vector stores by purpose.
  • GenAI and RAG: Usable generative AI needs access to internal knowledge. RAG connects language models with company documents while preserving model separation.
  • Data Governance: Trusted data products need clear rules and live quality signals. Data Governance and Observability define responsibility, standards and lineage.
  • Data Mesh: Scalable data use in large organisations needs domain ownership. Data Mesh treats data as products managed by the teams closest to each domain.
  • Tech Radar: Clear technology decisions need shared orientation. A Tech Radar evaluates trends, tools and methods systematically and keeps focus manageable.
  • Tech Debt: Sustained delivery capacity needs visible technical debt. Planned refactoring keeps maintenance, architecture and development velocity manageable.
  • OSS Strategy: Stable technical foundations need a deliberate OSS strategy. Contributions, dependencies and upstream responsibility make decisions traceable.
  • Blockchain and Web3: Decentralised trust needs tamper-resistant records. Blockchain and Web3 support transactions, data ownership and Self-Sovereign Identity.
  • Edge Computing: Fast IoT decisions need processing close to the source. Edge Computing moves analysis to machines, factory floors and mobile devices.
  • Privacy and Anonymity: Data insights and privacy can be engineered together. PETs and anonymisation techniques protect sensitive information during analysis.
  • Digital Ethics: Responsible digital systems need clear accountability. Digital ethics frames AI decisions around responsibility, discrimination and transparency.
  • Post-Quantum Cryptography: Long-lived confidential data needs crypto-agility today. Post-Quantum Cryptography plans migration from RSA and ECC to quantum-resistant standards.
  • AI Governance: Controllable AI use needs an operating model. AI Governance defines model, data, oversight and evidence rules as the basis for sovereign AI.
  • AI Agents and Agentic Systems: Acting AI systems need clear control. AI agents plan, call tools and observe results in a loop, turning chat into goal-oriented workflows.
  • LLMOps and MLOps: Production AI needs repeatable operations. LLMOps and MLOps version models and prompts, measure quality, and monitor cost, latency and drift.
  • Prompt and Context Engineering: Better model output often starts with better input. Prompt and Context Engineering shape instructions, retrieved context and tool schemas before fine-tuning.
  • Vector Databases and Embeddings: Semantic search needs meaning as vectors. Embeddings and vector databases provide the foundation for similarity search, RAG and semantic retrieval.
  • AI Evaluation and Guardrails: Reliable production AI needs measurement and runtime controls. Evaluation checks quality systematically, guardrails keep inputs and outputs in scope.
  • Fine-Tuning vs RAG vs Prompting: Efficient AI builds start with the right lever. Prompting, RAG and fine-tuning balance cost, freshness and control across rising levels.
  • Sovereign AI: AI and data sovereignty need compatible architecture. Sovereign AI runs open models on Swiss or on-premises infrastructure.
  • Language Models: Sound model choices need visible trade-offs. Language Models range from self-hostable open models to closed frontier models behind external APIs.
  • Workflow Automation and Data Flows: Useful automation starts with controlled data flows. Workflow Automation connects systems, sets boundaries and keeps human control explicit.
  • A-Team: Resilient digital operations need open, replaceable building blocks. The A-Team connects sign-on, collaboration, knowledge and AI with existing tools.