Research
My research formalizes the space between retrieval and reasoning: context compilation, memory runtimes, precision-aware context, governed agentic delivery, and the control planes that make AI usable at enterprise scale.
Foundational program
A Context Compilation research program: one foundational paper introduces the theory, followed by a three-paper trilogy on representation, runtime, and precision-aware optimization.
Open trilogyPublished papers
2026 / paper
This paper proposes that context in AI systems should not merely be retrieved or remembered — it should be compiled. Context Compilation Theory introduces a formal framework where context is selected, transformed, governed, optimized, and lowered into executable context packs for downstream models, agents, and interfaces. The paper defines Context IR as a portable intermediate representation, analogous to compiler IRs in traditional software, and presents CompileBench as a benchmark specification for measuring compilation quality rather than recall alone.
DOI: 10.5281/zenodo.19490060
2026 / paper
Paper 1 operationalizes Context Compilation Theory. It defines a four-level Context IR, a compiler-pass taxonomy for enterprise context pipelines, formalizes context graph breaks, and treats governance as intrinsic to compilation through policy-as-types and a portable Context ABI.
DOI: 10.5281/zenodo.19546798
2026 / paper
Paper 2 extends the trilogy from semantics into runtime behavior. It argues that compiled context should live like managed memory: evidence blocks enter working sets, locality governs movement, speculative execution prepares likely futures, context GC removes dead state, and provenance remains attached through lifecycle transitions.
DOI: 10.5281/zenodo.19546800
2026 / paper
Paper 3 turns optimization into a first-class concern. It defines a semantic precision ladder, a distortion model for compiled context, mixed-precision assembly strategies, and recovery-aware compression so systems can stay cheap until risk, policy, or task criticality demands higher fidelity.
DOI: 10.5281/zenodo.19546802
2026 / paper
This paper argues that the traditional SDLC cannot govern a world where AI agents produce engineering artifacts. Cybernetic Software Delivery (CSD) reframes delivery as a control system — a closed-loop model with humans, agents, tools, evaluations, and feedback. The primary unit of delivery becomes the governed run: a bounded, observable, policy-aware execution of delegated work that is fully reconstructable from storage. The paper introduces a nine-stage lifecycle, a five-category artifact taxonomy, 13 new metrics extending DORA, and a four-tier governance framework.
DOI: 10.5281/zenodo.19490008
Research themes
Layers
A research program for how AI systems should compile context across representation, runtime, and precision-aware optimization rather than treating retrieval as the entire stack.
Unifies fragmented approaches to RAG, memory, runtime discipline, and cost control under one formal context stack.
GitBranch
A governed lifecycle for engineering work produced by autonomous agents — treating delivery as a closed-loop control system.
Provides the operating model enterprises need when AI agents become first-class producers of engineering artifacts.
Cpu
Architecture patterns for systems where AI agents are first-class participants — scheduling, resource management, inter-agent communication, and governance.
Enterprises deploying agents need OS-level primitives, not just prompt wrappers.
Brain
How AI preserves continuity of cognition across changing data, models, tools, and experiences — from episodic memory to persistent context graphs.
Without durable memory, AI systems start every interaction from zero. Memory is the bridge between intelligence and usefulness.
TrendingUp
The cost structures, optimization strategies, and economic models that determine whether enterprise AI systems are sustainable at scale.
AI costs can spiral without visibility. Token economics is the CFO-level discipline of enterprise AI.
Shield
Governance, observability, policy enforcement, and trust calibration for enterprise AI — the management layer above models and agents.
Enterprise AI without governance is a liability. Control planes make AI auditable, explainable, and safe.
Where research connects
Essays translate the research program into operating implications for AI leaders and builders.
Industry publications connect technical shifts to model, vendor, infrastructure, and application-layer movement.
Guided paths for executives, builders, researchers, and industry watchers.