Hakamana is not a product pitch. It is a thesis about what happens when you build an enterprise AI knowledge platform that is Indigenous-led from the ground up. Not adapted, not culturally informed, not "inclusive." Led. Designed. Governed. By the communities whose knowledge it serves. Here is what we are building and why it matters beyond any single community.
The Scale of the Problem
Across Aotearoa and the Pacific, Indigenous communities hold vast knowledge repositories: oral histories, governance documents, Treaty records, land management data, genealogical archives, cultural practices, language resources. This knowledge exists in fragments across organisations, personal collections, institutional archives, and community memory.
The fragmentation is not accidental. It is the legacy of colonial knowledge systems that categorised, extracted, and scattered Indigenous knowledge across institutional boundaries designed by and for someone else.
The scale is significant. The corpus we are working with exceeds 80,000 knowledge assets across multiple iwi, hapu, and Pacific communities. Documents, transcripts, recordings, images, maps, governance records. Each asset with its own provenance, its own access protocols, and its own cultural significance.
80,000+
knowledge assets across Indigenous communities in the Hakamana corpus
The Thesis
The Hakamana thesis rests on three propositions:
1. Indigenous Knowledge Requires Indigenous Governance
Standard enterprise AI knowledge platforms treat all content as equal. Ingest it, chunk it, embed it, retrieve it. The governance model is access control: who can see what.
Indigenous knowledge requires a richer governance model. Some knowledge is for everyone. Some is for specific communities. Some is for specific roles within communities. Some is seasonal. Some is gender-specific. The access protocols are not simple permissions. They are cultural protocols that reflect relationships, responsibilities, and tikanga.
The Hakamana architecture encodes these protocols at the data layer, not the application layer. Every knowledge asset carries governance metadata that determines how it can be retrieved, who can access it, in what context, and with what obligations. The AI system does not override these protocols. It respects them as architectural constraints.
2. RAG Can Be Culturally Governed
Retrieval-Augmented Generation is the standard pattern for enterprise knowledge AI. Ask a question, retrieve relevant documents, generate an answer grounded in those documents.
The standard criticism of RAG for Indigenous knowledge is valid: it flattens cultural context, strips provenance, and presents knowledge without the relationships and protocols that give it meaning.
Our approach modifies the RAG pattern in three ways:
Governed retrieval. The retrieval step respects cultural access protocols. If a knowledge asset is restricted, it is not retrieved, regardless of its relevance to the query. The AI system cannot surface knowledge that the user does not have cultural authority to access.
Contextual generation. The generation step includes provenance and context. Answers do not just cite sources. They locate the knowledge within its cultural context: where it comes from, who holds it, what obligations accompany its use.
Audit trail. Every retrieval and generation event is logged with full provenance. Communities can see how their knowledge is being accessed, by whom, and for what purpose. This is not optional logging. It is an architectural requirement.
The standard RAG pattern treats knowledge as data to be retrieved. The architecture must reflect that difference, or it extracts value while claiming to serve the community.
Dr Tania Wolfgramm
Chief Research Officer
3. Enterprise-Grade and Values-Driven Are Not Contradictions
The technology industry treats "enterprise-grade" and "values-driven" as separate concerns. Enterprise-grade means scalable, reliable, secure. Values-driven means ethical, culturally responsive, community-centred. The Hakamana thesis says these are the same thing.
A knowledge platform that violates cultural protocols is not secure, regardless of its encryption standards. A platform that surfaces knowledge without provenance is not reliable, regardless of its uptime. A platform that does not serve the community's interests is not fit for purpose, regardless of its performance benchmarks.
Enterprise-grade, properly understood, includes the values that the enterprise (in this case, the community) holds. The architecture must be both technically rigorous and culturally grounded. Neither alone is sufficient.
The Architecture
Mak has led the technical architecture. The key design decisions:
NZ-sovereign infrastructure. All data processing and storage happens on New Zealand infrastructure. Non-negotiable. The knowledge does not leave the jurisdiction of the communities that own it.
Modular knowledge domains. Each community's knowledge exists in its own domain with its own governance rules. Domains can share knowledge between communities when both communities consent. The sharing protocols are explicit, auditable, and revocable.
Multi-model inference. Different knowledge types require different AI capabilities. Document analysis for archival materials. Speech-to-text for oral histories. Image analysis for visual records. The orchestration layer selects the right model for each task while maintaining governance constraints.
Community-controlled access. Each community manages its own access policies through governance interfaces designed for non-technical users. The governance is not delegated to a system administrator. It is held by the community's designated knowledge guardians.
Standard RAG treats the knowledge base as a pool of content; Hakamana treats it as a network of governed relationships. The engineering is harder, but the outcome is fundamentally different.
Mak Khan
Chief AI Officer
Why This Matters Beyond Indigenous Communities
The Hakamana approach addresses problems that are not unique to Indigenous knowledge:
Knowledge governance is a challenge for any organisation where different information has different access requirements, different owners, and different usage protocols. Legal firms, healthcare providers, financial institutions, government agencies: all of these need governed knowledge AI.
Provenance and trust are universal requirements. Any knowledge AI system that cannot explain where its answers come from and how much to trust them is incomplete.
Community sovereignty over data is an emerging requirement for many communities, not just Indigenous ones. Patient communities, professional communities, geographic communities: the principle that data about a community should be governed by that community is gaining traction globally.
The Hakamana thesis, built for and by Indigenous communities, is producing architecture patterns that have broad enterprise applicability. The most demanding governance requirements produce the most robust architectures. This is a pattern we have seen before, and it holds true here.
Hakamana is early. The architecture is proven at pilot scale. The 80,000+ asset corpus is being structured and governed. The community partnerships are in place. The work ahead is significant. But the thesis is clear: enterprise AI knowledge platforms can be Indigenous-led, enterprise-grade, and values-driven. Not as a compromise between competing priorities, but as an integrated design that treats cultural governance as an engineering requirement. That is the standard we are building toward.