Core Labs is an applied research laboratory operating at the intersection of artificial intelligence, physics-based simulation, cybersecurity, quantitative finance, energy systems, and engineering. We build autonomous, self-improving AI infrastructure that learns across disciplinary boundaries — surfacing structural patterns that no single field can reveal on its own, and converting them into deployable, verifiable systems.
Research Tracks
AI Specialists
Data Records
Autonomous
Representations learned in one domain measurably strengthen performance in adjacent ones. A model trained on quantitative risk analysis develops sharper security-audit capabilities — and the transfer runs both ways. We design curricula and evaluation protocols specifically to induce, measure, and exploit this cross-domain generalisation.
Our self-improving training loops run continuously — generating hypotheses, curating data, fine-tuning, and evaluating without a human in the loop. To date the platform has completed 48K+ autonomous learning sessions, each logged, scored, and fed back into the next training cycle as a closed experimental system.
The entire stack — training, inference, orchestration, and evaluation — runs on fully self-hosted, on-premises infrastructure. Research data never leaves the laboratory perimeter, and enterprise-grade guardrails enforce domain-appropriate safety policies across every vertical we operate in.
Rather than one monolithic generalist, we operate 56 domain-specialist AI agents — each distilled for depth in a specific discipline — coordinated by a central orchestration layer that decomposes tasks and routes each sub-problem to the most competent specialist.
Reproducibility is enforced cryptographically, not by convention. Every dataset, every training run, and every evaluation result is Merkle-verified, giving each artefact a complete, tamper-evident provenance chain from raw source data to deployed model weights.
We hold that scientific progress compounds when methods are inspectable. Our tooling, training methodologies, and trained models will be released to the research community — because independent replication and adversarial scrutiny are what turn results into knowledge.
The framework layer: every stage of the model lifecycle — training, alignment, safety, deployment, and agent execution — maps onto a dedicated NVIDIA component.
The operational layer: what the stack delivers in practice — measured, versioned, and continuously exercised by the autonomous training pipeline.
Autonomous training pipelines, multi-teacher knowledge distillation, large language model research, parameter-efficient fine-tuning, and frontier-class evaluation systems with rigorous statistical baselines.
Humanoid locomotion, dexterous manipulation, and sim-to-real transfer. Physics-accurate synthetic environments generate the scale and diversity of experience that real-world data collection cannot.
Static and dynamic vulnerability analysis, automated code auditing, applied cryptography, and reproducible security evaluation harnesses spanning entire software ecosystems.
Quantitative modelling, systemic risk analysis, agent-based market simulation, and macroeconomic forecasting driven by AI pattern recognition over heterogeneous time-series data.
Compute-resource scheduling and optimisation, energy-efficient training regimes, smart-grid load dynamics, and sustainability research for large-scale AI infrastructure.
Text-to-3D generation, procedural and parametric modelling, interactive real-time environments, and production-grade digital asset pipelines for industrial applications.
32 active research tracks spanning computer science, mathematics, physics, and engineering — curated in collaboration with curricula and open literature from leading universities, with an emphasis on interdisciplinary synthesis.
Low-latency real-time AI interaction, procedural content generation, agent-driven experiences, and structured human–AI collaboration frameworks with measurable task outcomes.
Large-scale dataset curation and deduplication, statistical analysis, signal processing, and multi-modal learning across structured, unstructured, and streaming data sources.
Sets the laboratory's research agenda and long-range technical bets, cultivates institutional and industry partnerships, and directs the go-to-market strategy for the Core Labs AI platform and its commercial products.
Owns the full NVIDIA stack integration end to end — from distributed training reliability and inference serving to on-premises infrastructure operations, observability, and the security posture of the entire platform.
Our self-training AI platform and flagship product: 56 specialist agents orchestrated over 371K curated training records, currently ranked #4 globally in domain evaluations. Meet Molly →
Core Labs is established in Panama. The initial research infrastructure comes online and the first multi-node compute cluster is assembled — laying the physical and organisational groundwork for fully self-hosted, privacy-first AI research.
The autonomous training platform launches on NVIDIA Grace-Hopper. 32 academic research tracks are activated and 371K cross-domain training records are curated, deduplicated, and Merkle-verified — the substrate for every subsequent training cycle.
The enterprise partnership is activated, completing full-stack integration across NeMo, NIM, Guardrails, NemoClaw, Isaac Sim, and the NeMo Agent Toolkit — unifying training, inference, safety, and agent execution under a single supported platform.
Commercial launch of the autonomous AI platform at iamolly.ai, bringing Core Labs' domain-specialist agent architecture to enterprise and research customers as a production product.
New research verticals open — energy optimisation, interactive systems, and industrial simulation — alongside multi-GPU cluster scaling to support the growing volume of autonomous training and evaluation workloads.
Loading…
Core Labs builds at the intersection of AI, simulation, security, finance, energy, and engineering — on self-hosted infrastructure, with verifiable methods. Our flagship product, Molly AI, brings that research directly to enterprise and research teams.