YARA Research is an independent AI research and engineering initiative exploring the foundations, architectures and emerging possibilities of intelligent systems.
YARA Research investigates the fundamental principles and practical engineering of intelligent systems. We approach AI through the lens of rigorous research: forming hypotheses, designing experiments, publishing findings and releasing reproducible implementations.
Our independence from commercial constraints allows us to explore questions that matter for the long-term development of AI, not only the ones that are immediately profitable.
Every investigation begins with evidence, not intuition. We examine the literature, establish baselines, and design experiments before forming conclusions.
We validate hypotheses through reproducible experiments. Findings are reported with their limitations, negative results and open questions intact.
Research that cannot be independently verified is not complete. We publish methodologies, code and datasets alongside our findings.
YARA Research investigates AI through six interlocking programmes, each exploring a distinct dimension of intelligent systems.
Investigating the nature, architectures and mechanisms of machine intelligence, from reasoning and inference to model architecture and knowledge representation.
Investigating autonomous and non-autonomous approaches to intelligent task execution, comparing agent architectures, agentless alternatives, and hybrid strategies.
Researching how AI transforms software engineering and development infrastructure, from coding agents to AI-native software development lifecycles.
Making AI more efficient, accessible and independent of centralised infrastructure, through local inference, model optimisation and resource-aware computing.
Understanding how intelligent systems acquire, retain, retrieve and use information, from context engineering to persistent knowledge architectures.
Ensuring intelligent systems are reliable, secure, measurable and accountable, through rigorous evaluation, guardrails and responsible AI practices.
Three initial research projects investigating the architecture of next-generation intelligent systems. Each project follows the YARA research lifecycle: from hypothesis through experimentation to reproducible publication.
Investigating Agentless and Hybrid Intelligence Architectures
Exploring the boundaries between deterministic, agent-based and hybrid approaches to intelligent problem solving. Can we achieve reliable and efficient task execution through deterministic workflows, structured reasoning and selective autonomy?
Researching Modular Execution Architectures for Intelligent Systems
Investigating the architecture and engineering principles behind lightweight, modular intelligent runtimes. Can we develop a modular AI runtime that provides essential orchestration without depending on heavyweight agent frameworks?
Exploring Small Models, Local Inference and Context Efficiency
Investigating how small language models, local inference, retrieval and intelligent context selection can be combined to deliver efficient AI systems on resource-constrained hardware.
Every YARA Research project passes through a defined lifecycle: from initial idea through experimentation, independent review and public release. Research governance ensures quality, reproducibility and integrity at each stage.
Note: A project may return to experimentation if results are not reproducible. Negative findings and limitations are always published alongside positive results.
YARA Research follows a rigorous publication lifecycle. Research papers, technical reports and working papers will appear here as projects progress through experimentation, review and publication approval.
We will not publish preliminary claims as completed research. Quality and reproducibility take precedence over velocity.
We publish code, benchmarks, and experimental environments alongside our research. True scientific progress requires that every claim can be independently tested, replicated, and inspected.
Comparative evaluation harness for deterministic workflows vs autonomous agent loops on structured enterprise tasks.
Experimental modular execution runtime providing state-machine orchestration, permission boundaries, and context efficiency.
Reference implementations for quantized local model routing, adaptive context selection, and constrained decoding.
YARA Research is designed to operate as a distributed, globally connected research community. Researchers, engineers, students and independent contributors are welcome, regardless of geography or institutional affiliation.
Follow active research projects, review literature and provide feedback on published findings. Open to students and enthusiasts interested in AI research.
Conduct experiments, implement prototypes, review code and contribute to open-source research tools. For engineers and researchers with relevant technical background.
Take ownership of a research workstream within an active project. Requires sustained commitment and results in co-authorship of published outputs.
Lead research direction within one of the six programmes. Senior role requiring deep domain expertise and long-term commitment.
Provide independent technical and methodological review of research before publication. For experienced researchers in relevant domains.
Contribution roles are currently in planning. These are voluntary collaboration roles, not employment positions. When participation opens, clear agreements covering IP ownership, attribution, licensing and contribution expectations will be published. Follow our GitHub organisation for updates.
We explore the questions that shape the next generation of intelligent systems. Independent research. Open science. Reproducible knowledge.