Building AI-Ready Nations through Digital Public Infrastructure
AI does not require a new framework — it requires the right integration with existing DPI foundations. A practical approach to embedding modular, governable intelligence into public service delivery.
The value of AI emerges not from its autonomy
but from how it integrates with DPI
This paper presents the DPI-AI Framework — a practical way to think about how Artificial Intelligence can be integrated into public digital systems through Digital Public Infrastructure. Written at a moment when AI capabilities are advancing rapidly while many governments still grapple with fragmented systems and legacy architectures.
Rather than proposing AI as a standalone transformation, the framework explores how existing DPI foundations can provide structure and coherence for AI in the public sector. It deliberately avoids prescribing specific technologies or models. Instead, it offers a shared mental model.
This framework does not assume AI should be prioritised ahead of foundational DPI work. In many contexts, the immediate focus remains on clean registries, reliable APIs, and minimal institutional capacity.
The paper is intended for digital leaders in government, development partners, donors, and practitioners who are shaping national and global digital agendas.
At its core, government is a system of:
AI Blocks can read and write registries. DPI Workflows encode and execute rules. Public Agents interpret citizen intent and trigger the right combination of both.
Three interrelated elements
Together they describe how intelligence can be embedded into public service delivery while preserving interoperability, accountability, inclusion, and sovereignty.
Modular AI Capability
Callable functions that encapsulate a single AI capability — the ingredients of any AI-enabled service. Minimalist, composable, reusable, and governable.
Sector-specific: eligibility verify, identity verify, clinical decision support
The Orchestration Layer
The recipe that coordinates AI Blocks with DPI systems, policy rules, data flows, and human oversight. The workflow is the locus of control — AI operates within it, never outside it.
AI-Enabled Interfaces
Constrained interfaces — not decision-makers — that interact with citizens and public servants. They activate DPI Workflows and interpret citizen intent while remaining accountable extensions of government.
6 Principles · 7 Patterns
The principles explain why the framework is designed this way. The patterns show how to apply it in practice.
Interoperability First
AI Blocks connect through open standards — not proprietary integrations. Any block can be swapped without rebuilding the workflow.
DPI as the Source of Truth
AI outputs inform decisions; registries remain authoritative. No AI block directly modifies authoritative records without a governance step.
Minimal Footprint
Each AI Block does one thing well. Complexity is managed at the workflow level, not embedded in blocks.
Human Oversight by Default
Any low-confidence output, exception, or safeguard violation routes to a human. Automation is earned, not assumed.
Sovereignty & Replaceability
Model provenance, hosting options (sovereign cloud / on-prem / external API), and replaceability conditions are declared in every block spec.
Govern Data, Not Just Models
Data provenance, consent, and stewardship are foundational. A government cannot train a trustworthy model on ungoverned data.
The +1 Approach: Governments don't need to replace legacy systems to become AI-ready. Layer new capabilities alongside what exists — wrapping legacy systems with interoperable APIs and DPI blocks. Every registry cleaned, every API published, every governance process formalised is a direct enabler of AI-ready public infrastructure.
Making the vision tangible
Five sector archetypes showing how AI Blocks, DPI Workflows, and Public Agents combine to deliver real services.
eligibility_verify() checks against social registry and policy rulesdocument_extract() processes supporting documents (OCR)ocr_extract() processes physical birth notification formsidentity_verify() cross-checks parents against civil registryspeech_to_text() transcribes in local languagetranslate() normalises to administrative languagestructured_extract() pulls name, location, crop type, land sizepersonalized_tutor() generates adaptive content in local languageprogress_assess() tracks mastery and identifies gapsarea_classify() identifies affected population from registrieseligibility_verify() rapid-matches against social protection registryfraud_detect() screens for duplicate or anomalous claimsFrom architecture to deployment
9 sequential steps — each producing a concrete delivery output. Later steps rely on decisions and artifacts produced earlier. Skipping steps typically results in weak governance, fragile automation, or unclear accountability.
Identify the Use Case & User Journey
Start with one narrowly defined service journey that can be completed end-to-end. Map it in plain language before any technical work begins.
Identify AI Blocks, DPI Dependencies & DPGs
For each step in the journey, identify the minimum set of AI capabilities required. Each capability must be callable as a block, invoked only through the workflow.
Define Governance, Safeguards & Human-in-the-Loop
Before automation: owning ministry, legal mandate, consent requirements, escalation rules, audit and retention policies. Human oversight is not optional.
Create the DPI Workflow
Translate the user journey into a machine-readable DPI Workflow. Specify ordered steps, safeguards between steps, fallback logic, and authoritative outputs.
Define the Public Agent
Configure as a constrained interface. Supports local languages, can activate only approved workflows, cannot access data without consent, always offers human escalation.
Define Metrics & Operational Signals
Metrics must be defined before launch: success/failure rates per step, escalation frequency, language coverage, confidence scores, end-to-end completion time.
Implement the Service
Stand up the actual system. Code-based (APIs/SDKs) or No-Code/Low-Code (OpenFn, N8N). Both paths produce the same outcome. Start in supervised, human-in-the-loop mode.
Test & Validate
Synthetic end-to-end tests, AI Block accuracy validation against declared thresholds, safeguard validation, and human-in-the-loop dry run. No citizen contact before validation.
Fine-Tune & Improve Iteratively
Adjust confidence thresholds, expand language coverage, reduce manual escalation where safe, improve data mappings. All improvements via blocks and workflows — not ad-hoc logic.
See the framework in action
Watch a live end-to-end demo of a farmer requesting a government benefit via WhatsApp — with every AI Block, DPI system, and governance step visible in real time. Then assess your readiness, explore scenarios, and generate a roadmap.
Open the Sandbox →Read, cite, and implement
DPI-AI Framework
The full framework paper: executive summary, architecture, use cases, challenges, glossary, and references.
From Architecture to Deployment
Step-by-step playbook with YAML templates, AI Block specifications, and governance checklists for running a pilot.
The DPI-AI Framework
Extended version of the framework paper with detailed architecture diagrams, patterns, and implementation guidance.
Centre for Digital Public Infrastructure
CDPI works to accelerate the adoption of DPI globally, providing technical assistance, convening, and knowledge resources.
Visit cdpi.dev →Citation
CDPI Abadie, D. (2026). DPI-AI Framework: Building AI-Ready Nations through Digital Public Infrastructure.
https://digitalpublicinfrastructure.ai →Creative Commons CC BY 4.0
This work is licensed under Creative Commons Attribution 4.0 International. Free to share and adapt with attribution.
View license →