Loading...
Loading...
Our research team investigates intelligence systems at the system level, so that we can optimize AI-empowered value creation in every aspect of our lives.
Research Themes
We investigate the latent plasticity of foundation models to enhance their generalizability in data-sparse environments. Our work focuses on increasing the operational value of models and agents through systematic improvements in training, tooling, and evaluation frameworks.
We build systems and applications that orchestrate seamless human-agent and agent-agent collaboration, abstracting technical complexity into applications that render sophisticated systems legible and operable by non-specialists at every node of the value chain.
We quantify intelligence to measure the fundamental capacity of value networks. With novel metrics and modeling methodologies, we isolate the high-fidelity cognitive signal from the stochastic noise of models.
Publications
The Vicunous Harness Runtime is a specialized execution environment designed to securely productionize autonomous AI agent harnesses. By decoupling stateful harness logic from execution infrastructure, the runtime allows domain experts to perform harness design and engineering while seamlessly enforcing hardware-level compute isolation and zero-trust credential management in multi-tenant production deployments.
Adaptive AgencyA benchmark comparing general-purpose models, cross-domain methodological models, and traditional machine learning for funding tasks, with weighted scoring based on domain relevance and a practical ROI lens.
Intelligence TheoryA framework for understanding, measuring, and enhancing adaptive capacity in complex systems.
Adaptive AgencyBenchmarking general-purpose LLMs against domain-specific specialists requires more than models; it requires the right testing environment: curated datasets, a repeatable ETL pipeline, and feature engineering calibrated to the temporal rhythms of financial data. Here we detail the data infrastructure and design decisions behind it.
LaymanizationWe present the V Stack, a seven-layer operational architecture for building AI-native, Agent-first, Adaptable Intelligence (A3I). Modeled after the OSI reference model, each layer defines a clear boundary, from persistent organizational wisdom at L1 to laymanized interfaces at L7, structuring how autonomous agents execute full business operations at Vicunous.
Adaptive AgencyA taxonomy of the foundation model landscape—with a focus on Domain-Specific Foundation Models (DSFMs)—and how it informs model selection for funding applications.
LaymanizationA systems-level thesis for how technologies become usable by non-experts, how infrastructure compounds adoption, and why intelligence systems will reshape value flow networks.