Representation
Native Meaning Units
A unit of semantic structure intended to capture meaning directly, rather than as a statistical by-product of token sequences.
Independent Research — Architecture & Epistemics
NINMENI Research Lab
Independent Indonesia-native epistemic AI architecture research.
NINMENI investigates non-Transformer approaches to language understanding through Native Meaning Units, grounded world-state updates, contradiction and gap tracking, inquiry generation, MRO selection, and controlled rendering.
Research question
Can language understanding be modeled as the formation, stabilization, grounding, and revision of meaning structures, rather than direct next-token prediction?
NINMENI explores this question through an Indonesia-native architecture built around meaning units, world-state, contradiction awareness, inquiry, and controlled rendering.
Research focus
State
Grounded World-State
A maintained representation of entities, relations, and edges, updated as information is ingested, rather than re-derived at each step.
Consistency
Contradiction & Gap Tracking
Explicit detection of conflicting or missing information within the world-state, as a first-class part of the architecture.
Direction
Inquiry Generation
Formulating questions the system needs answered to resolve gaps or contradictions, rather than producing output regardless of uncertainty.
Output
Controlled Rendering
Output is rendered from structured meaning and epistemic status, rather than from surface fluency alone.
What NINMENI is
NINMENI is an experimental architecture research project exploring whether language understanding can be represented as structured epistemic formation rather than direct next-token prediction.
It is being developed independently from Indonesia, using the Indonesian language and public-domain structured data as the primary research environment.
What NINMENI is not
- Not a Transformer wrapper
- Not a fine-tuned LLM
- Not a chatbot interface over another model
- Not a model extraction or distillation project
- Not an AGI claim
- Not production-ready
Current research status
Internal diagnostic milestones — not benchmark results.
Accepted
Corpus Freeze V0
Pass
Micro-Training V0 — diagnostic pass
Accepted
Scope-A year-scoped value edge
13,553
Native Meaning Units ingested
2,334
World-state nodes
1,310
World-state edges
820
Grounded edges
490
Speculative edges
746
Gaps & inquiries
Pass
Deterministic replay
50,000
Next diagnostic target — NMU
Technical evidence
NINMENI is currently documented through internal diagnostic gates, corpus freeze reports, micro-ingestion summaries, and scoped value-edge evaluations. A sanitized research-preview repository can be prepared for technical review when appropriate.
The public repository may not always reflect the latest internal architecture, because some implementation details are intentionally withheld while the research is still evolving.
Responsible research use
NINMENI is developed as an independent research project with responsible-use boundaries. External AI systems may be used for review, coding assistance, evaluation design, and documentation support, but not as hidden inference engines, distillation sources, or extraction targets.