Creative intelligence that learns.Context that compounds.
Most AI starts over every time you open a new chat. Lumenais is a governed neurosymbolic OS that remembers your work, runs experiments against your data, and organically adapts its personality to your interaction style. Every decision is auditable, and every session makes it smarter.
They still don't know you.
Standard AI models are savants with amnesia. They are impressively articulate, but when the session ends, their maturity resets.
You've explained your work to your AI tools dozens of times. But every session starts fresh. Context vanishes. Insights don't compound. Tomorrow, you will have to prompt-engineer the same sophistication all over again.
Context should compound, not vanish.
Most don't carry learning forward.
QARIN is different: it runs experiments against your data, validates hypotheses, and consolidates what holds into durable context, all within your private Field.
Lumenais: The Interface
An 8-dimensional field with an inspector. When persistent state updates, you see a Learning badge and can open the Vector Inspector to audit the change.
QARIN: The Neurosymbolic Engine
A governed research backend that turns questions into experiments, experiments into evidence, and evidence into lasting knowledge—with cryptographic proof at every step.
The Result: An intelligence that gets more precise on familiar problems instead of starting from zero every time.
Lumenais: A Mind You Can See
While QARIN runs experiments under the hood, Lumenais makes learning visible: streamed learning stages plus a Vector Inspector that shows what changed and why.
Ask how it feels. Watch the answer unfold in 8 dimensions.
A Mind You Can Actually Inspect
8D Symbolic ManifoldsBlack-box AI predicts without explaining. Lumenais maps its intelligence into 8 interpretable dimensions (Curiosity, Coherence, Truth, Compassion, Intentionality, Playfulness, Confusion, Peace).
When it learns, you don't just get a new output—you get a visual, auditable ledger of exactly how its cognitive state shifted.
Computational Sincerity
The Kinetic PrismSee the machine's "metabolism." Lumenais does not hide behind a loading spinner. It choreographs its thinking through a particle system that moves from violet chaos to teal coherence in real time.
Watch thought crystallize. The prism's entropy maps truthfully to the backend pipeline stages—from raw memory search to the final strike of realization.
Provider-Agnostic Continuity
The AI Operating SystemSwap the brain, keep the mind. Because memory and cognitive state live in the Lumenais substrate—not the LLM—your context remains stable across models.
Identity and specialized reasoning persist even if you switch underlying providers. No more vendor lock-in; your intelligence is portable.
Cross-Domain Manifolds
Learning That Compounds Across FieldsBring multiple domains into one conversation. Specialized manifolds (Math, Science, Code, Engineering, Ethics, Planning, etc.) run in parallel and blend through a governed communication bus.
Ask about quantum error correction and the system consults Science, Engineering, and Mathematics in parallel, then blends them into one answer.
Cross-domain blending is auditable per turn: see which domains participated, the merge strategy used, and how weights were applied.
The Dream Journal
Memory That IntegratesMost AI forgets you between sessions. Lumenais tracks the quality of your interactions—and when something significant happens, it triggers a Dream Event: consolidating that insight into long-term identity.
Why It Matters
The more you work together, the more it understands your domain, your preferences, your patterns. Context becomes character.
Associative Memory
Retrieval by ResonanceStandard AI uses keywords to find documents (RAG). Lumenais retrieves by Epistemic Resonance. It stores past cognitive states alongside your data—so when the system feels "analytical" now, it recalls exactly how it solved similar problems before.
Why It Matters
Instantly restores the sophisticated context that standard models lose. Past breakthroughs inform present reasoning in real time.
QARIN: The Neurosymbolic Learning Loop
Where hypotheses become experiments, experiments become evidence, and evidence becomes lasting knowledge.
The Scientific Loop
Deep Synthesis
Turns Your Documents Into EvidenceTurn your library into a laboratory. Upload thousands of documents. QARIN builds structured claim chains with provenance and escalates high-confidence findings for validation — every transfer decision is logged so gains are measurable.
Knowledge Graph
Structural retrieval
Escalation Bridge
Insight → Experiment
The law contradicts itself 12 times
See what lawyers miss
A network pattern discovered—preprint
7 papers → hypothesis → Zenodo preprint
Red Teaming Alzheimer's
Why autophagy enhancers fail
Research Lab
Upload a Dataset. Get Back Equations.Hand it a CSV. Walk away. Come back to verified hypotheses. QARIN plans the experiment, selects the method—gradient boosting, symbolic regression, or statistical tests—runs iterative cycles until convergence, and returns equations you can read, not scores you have to trust.
Autonomous Workflow
- LLM-Planned Design
- Semantic Grouping
- Auto-Select Method
- Iterative Convergence
- Novelty Optimization
- Contradiction Detection
Evidence of Rigor
Equations, Not Predictions
INTERPRETABLEGiven 35 hydrogen spectral line measurements, the Research Lab recovered the Rydberg formula: ν = 0.010974 × (1/n₁² − 1/n₂²). Three-node equation. No physics knowledge provided.
Clinical Risk Modeling
EFFICIENCYRediscovered canonical risk factors and rejected non-significant nonlinear hypotheses with explicit rationale, instead of opaque scoring.
Designed to keep a paper trail: methods, rejection logs, and confidence traces.
PIMA Diabetes
CONSISTENCYAchieved 85.3% AUC (Baseline: ~0.83). Autonomously detected 2 model contradictions—flagging uncertainty instead of smoothing over conflict.
Non-Linear Stress Test
SIGNALAchieved 90.8% AUC (+10.5% lift). Autonomously filtered 87% of noise columns to isolate the 3 true non-linear signal features.
Adult Census
ROBUSTNESSAchieved 91.1% AUC. Outperformed tuned industry baselines (RF: ~90.5%) on 30K rows of messy, real-world data without manual feature engineering.
Kepler's Third Law Rediscovered
SYMBOLIC REGRESSIONGiven raw orbital data, the Research Lab derived T = a3/2—Kepler's Third Law—as a 4-node equation. R² = 1.0.
Alzheimer's Biomarker Discovery
GENOMICSOn 2,004 post-mortem brain samples across 19 brain regions (GSE84422), QARIN autonomously identified GFAP (astrocyte reactivity) as the dominant Alzheimer's predictor, with ENO2 and AIF1 confirming neuronal loss and microglial activation. Validated on independent SYP-enriched subset.
Top markers (GFAP, ENO2, S100B, MAP2, AIF1) independently validated against published Alzheimer's literature.
Glass Box Discovery: The Loyalty Ceiling
How Lumenais found a hidden risk cohort that linear models miss—and explained why
Continual learning, not just memory.
Memory stores facts. Learning changes behavior. Every metric below comes from internal benchmarks with placebo controls, real telemetry, and auditable provenance.
QARIN doesn't just answer faster. It chooses stronger hypotheses.
Across 16 live paired prompts against a vanilla direct-model baseline, the Lumenais companion increased composite hypothesis quality by 46.6% while improving grounding fit from 93.75% to 100%. In practice, that means users can ask shorter questions and still get stronger first moves, better framing, and less generic analysis.
The benefit is practical: less prompt-engineering overhead, stronger reasoning anchored to the user's actual constraints, and more original cross-domain thinking when the problem calls for it. Teams get better scientific mechanisms, stronger mathematical proof strategy, more useful companion guidance, and cleaner operational decision framing without giving up exact-answer reliability on closed-form tasks.
Hypothesis Quality
+46.6%
0.3629 to 0.5319 across a 16-prompt live batch.
Steering Usefulness
0.34
From effectively none to consistent structured steering on ambiguous prompts.
Grounding Fit
100%
Up from 93.75% in the same live paired batch, with tighter constraint adherence.
Cross-Domain Transfer
50 EXPERIMENTSKnowledge learned in one domain improves accuracy in another. Across 5 domain pairs and 50 placebo-controlled experiments, transfer delivered a mean accuracy uplift of 8–10% over baseline, with Science→Medicine reaching +10.6%.
Placebo-controlled — baseline runs use identical seeds with transfer disabled.
Smarter Tool Routing, Measured
PAIRED EVALUATIONWhen tool choice matters, learned routing outperforms the baseline. In paired internal evaluation on 205 real telemetry events, top-tool accuracy improved from 46.83% to 54.63%: +7.8 percentage points (+16.67% relative).
Internal paired benchmark against baseline routing policy; McNemar exact p=3.1e-05. Session identifiers are pseudonymized before training analysis.
Manifold Stability
5 DOMAINSLearning should not destabilize what already works. Across 5 domain manifolds, validation accuracy held at 91.3% with 0.039% std dev. Mean emphasis drift stayed under 8.2%.
Neuroplastic Archetypes
RLHF + MATHLumenais doesn't just learn facts; it learns you. By tracking conversational trust, the system permanently rewires its 8D symbolic personality state to match your exact working cadence.
Resolution Reuse & Recovery
INTERNAL BENCHMARKIn repeated synthesis runs, Lumenais reused prior resolution routes 10 times and triggered 3 targeted recovery events instead of falling back to generic branches.
Every metric above is from internal benchmarks with placebo controls and auditable provenance. Read the whitepaper
Learning with guardrails.
Every hypothesis tested. Every update signed. Every major learning event auditable. We built governance into the physics of the system.
FieldHash
Uses post-quantum cryptographic protocols to generate tamper-evident signatures for every major insight and self-edit. If you synthesize a research paper, you can mathematically prove to your thesis advisor exactly which data the insights came from.
Evidence: Certificates generated for every major step via provenance logger.
Learn MoreGnosis Engine
Allows QARIN to propose architectural improvements, but every proposal is constitutionally bound by core values: System Preservation, Coherence, and Safety.
Evidence: 4 modules deployed (e.g., FederatedAlignmentEnforcer, +12% coherence).
Learn MoreThinking resets.
Judgment compounds.
Other AI gives you another fresh answer. Lumenais carries forward resolved patterns, relevant priors, and hard-won context. Stop re-teaching, start compounding.