Frontier reasoning-systemen voor soevereine AI.
Supernova is de researchziel van NeuroCluster — benchmark-gedreven, anti-hype, en gericht op efficiënte, verifieerbare reasoning die enterprises op soevereine infrastructuur kunnen deployen.
Huidige AI is gecentraliseerd, ondoorzichtig en moeilijk te besturen. Supernova onderzoekt architecturen waarbij reasoning expliciet, verifieerbaar en deploybaar is op Europese hardware — met claims ondersteund door publieke benchmarks.
Supernova & Geitje 2
Frontier reasoning en een Nederlandse taalfundering — hoe onze modellen samenwerken op het platform.
Onderzoeksgebieden
Zes onderling verbonden onderzoekslijnen — benchmark-gedreven en ontworpen voor soevereine deployment.
Hierarchical Reasoning
Multi-step inference architectures that decompose complex problems into verifiable stages.
Verifiable AI
Reasoning traces, formal verification, and theorem-proving experiments with Lean.
Agentic Systems
Multi-agent coordination with governance boundaries and auditable tool execution.
Memory Systems
Persistent context, retrieval-augmented reasoning, and long-horizon task memory.
AI Governance
Policy-aware research aligned with enterprise deployment and EU regulatory requirements.
Efficient Inference
Sample efficiency, compute budgets, and energy-per-correct-answer metrics.
Publicaties
Supernova Whitepaper v0.1
Post-transformer reasoning research — problem statement, hypothesis, and benchmark strategy.
LeesSovereign AI Manifesto
Why European organisations need sovereign infrastructure, verifiable reasoning, and governance.
LeesBenchmark Methodology
Reproducible evaluation harness for ARC-AGI, planning, and theorem-proving tasks.
LeesBenchmark suite
Research roadmap
v0 Research
2024–2025Architecture exploration, thesis formulation, initial ARC-AGI baselines.
Benchmark Phase
2025Public leaderboard submissions, reproducible eval harness open-sourced.
Memory Systems
2025–2026Long-horizon context, retrieval integration, agent memory primitives.
Reasoning Engine
2026Hierarchical and latent reasoning models with verifier-based training.
Multi-Agent Orchestration
2026+Governed multi-agent mesh integrated with enterprise platform runtime.
Verder verkennen
Efficient verifiable reasoning systems.
Supernova investigates compact reasoning architectures that may complement or outperform transformer-based systems on structured reasoning tasks — with a focus on efficiency, verifiability, and sovereign deployment.
Research / ThesisProblem, hypothesis, and open questions.
Transformers dominate general language tasks but struggle on structured reasoning with efficiency guarantees. Supernova explores architectures where reasoning is explicit, verifiable, and deployable on sovereign hardware.
Research / BenchmarksBenchmarks before claims.
We publish benchmark results with full methodology — parameter count, training compute, energy use, and accuracy — before any product positioning.
Research / PapersPublications and experiment logs.
Research outputs publish as preprints, technical reports, and reproducible experiment logs — separate from product marketing.