Mainstream AI
Model
Product
Deployment
Adaptation
A systems-level framework for AI
AI, thought differently.
Constraint-native. Non-extractive. Viable beyond the cloud.
LowRes.AI designs from the realities that mainstream AI assumes away: limited connectivity, compute, literacy, data, technical support and institutional capacity.
Start with reality01 / 11
Start with the operating environment
A system is only intelligent if the whole system can work where it is needed. LowRes.AI treats six interdependent dimensions as design inputs from the beginning.
01Connectivity
Whole-System Viability
02Compute + Power
03Data + Knowledge
04Interaction
05Operational Capacity
06Institutional Capacity
02 / 11
The constraint-native shift
The order of design matters. Instead of adapting a finished AI product to constrained settings, begin with the environment and the outcome people actually need.
LowRes.AI
Operating environment
Required outcome
Viable system
Appropriate AI
03 / 11
More than a language problem
Better support for low-resource languages is essential. It is not, by itself, enough.
A model may understand a language and still fail the people who speak it. Weak connectivity, limited electricity, low literacy, absent technical support, recurring subscriptions and inappropriate governance can make an otherwise capable system unusable.
Non-extractive by designIndigenous and community knowledge should not simply become another dataset to scrape. Knowledge may need to remain local, attributable, governed by its holders and outside model training.
04 / 11
Beyond the subscription model
Access assumptions shape architecture. A viable system reduces the number of dependencies that must keep working forever.
Default stack
accountemailcloudsubscriptioncontinuous connectivity
VERSUS
Constraint-native stack
shared accesslocal infrastructureoffline capabilitylocal or institutional ownershipminimal recurring dependency
05 / 11
What qualifies as LowRes.AI?
Multiple constraints are first-class design inputs.
Those constraints materially change the architecture.
The system has a strategy for failure of mainstream assumptions.
06 / 11
Design principles
Not a checklist of features. A discipline for deciding what the system should—and should not—be.
- 01Constraint-first
- 02Minimum sufficient intelligence
- 03Graceful degradation
- 04Human-reality interfaces
- 05Bounded and attributable knowledge
- 06Lifecycle viability
- 07Human and local agency
07 / 11
First reference implementation
SILAC
A community-controlled knowledge environment for Indigenous communicators.
Its initial implementation focuses on climate resilience, biodiversity, forests and agriculture—supporting local interpretation and useful communication without dissolving knowledge into an opaque training corpus.
Trusted external knowledgeattributed
Community-governed Indigenous knowledgecontrolled
Local interpretation
Validation
Useful communication + action
SILAC does not silently merge Indigenous knowledge into a training dataset. Provenance and local control remain visible throughout the system.
Explore SILAC →08 / 11
One framework, many domains
The constraints differ. The discipline of whole-system viability travels.
Climate resilience
Agriculture
Health
Humanitarian response
Education
Public services
09 / 11
Why LowRes.AI?
Because the people most poorly served by mainstream digital systems should not have to wait for ideal infrastructure.
10+years of Audiopedia Foundation practice designing accessible knowledge systems where literacy, connectivity, language and devices cannot be assumed.
“LowRes.AI did not begin with an abstract question about what AI might become. It began with years of practical work asking what knowledge systems must do when the usual assumptions fail. SILAC is the first implementation of that thinking.”
11 / 11
Notes from the field
Ideas, tensions and lessons emerging as the framework meets practice.
Why I Started LowRes.AI
From accessible knowledge to constraint-native intelligence
Researchers · Implementers · Funders · Organisations
Let’s compare notes.
If you are working on AI in constrained environments—or questioning the assumptions built into today’s systems—we would like to hear from you.
Start a conversation