ENGINEERING BLOG
Technical research and engineering insights from the team building the operating system for responsible AI operations.
121 articles · Published by MARIA OS
Every decision must cite its source — how MARIA Meeting AI eliminates hallucinated minutes through segment-level evidence linking
Traditional meeting minutes suffer from a fundamental trust problem: the reader cannot verify whether a recorded decision actually occurred in the meeting or was interpolated by the note-taker. MARIA Meeting AI solves this by enforcing mandatory evidence linking — every decision, action item, and summary section must reference specific transcript segments as evidence. This paper formalizes the evidence-linking constraint, presents the incremental summarization algorithm that generates minutes every 15 seconds during live meetings, and proves that the citation coverage metric converges to completeness as transcript length increases. In evaluated Japanese business meetings, the system achieved 94% citation coverage with zero hallucinated decisions.
Dual-model anomaly detection, threshold engineering, gate integration, and real-time stability monitoring for autonomous agent systems
The Doctor system in MARIA OS implements organizational metacognition through dual-model anomaly detection, combining Isolation Forest for structural outlier detection and an Autoencoder for continuous deviation measurement. We detail the combined score A_combined = alpha * s(x) + (1 - alpha) * sigma(epsilon(x)), threshold design (soft throttle at 0.85, hard freeze at 0.92), and Gate Engine integration for dynamic governance-density control. We also define a stability guard that monitors lambda_max(A_t) < 1 - D_t in real time, where A_t is the operational influence matrix. Operational results show F1 = 0.94, mean detection latency of 2.3 decision cycles, and 99.7% prevention of cascading failures.
From keyword detection to action-level control: a formal shift that recasts AI routing from text classification to governance-aware execution control
Traditional AI routers treat routing as text classification: extract keywords, map to categories, and dispatch handlers. For enterprise-grade agentic systems, this approach is often insufficient. We formalize the Action Router as a function R: (Context × Intent × State) → Action, replacing the naive R: Input → Category mapping. The Action Router integrates with the MARIA OS Gate Engine so responsibility is enforced at routing time, not retrofitted afterward. We formalize the action space, define precondition-effect semantics for routable actions, derive routing cost over feasible actions, and show in simulation that action-level routing reduces misrouting by 67%, cuts responsibility-attribution failures by 94%, and achieves 3.2x lower latency than semantic-similarity routing on enterprise decision workloads.
How 111 agents across 10 roles self-organize, specialize, and form emergent hierarchies in the AGORA-100 simulation
We analyze role-specialization dynamics in Planet 100 (AGORA-100), a 111-agent governance cluster operating under the MARIA OS coordinate system. Using entropy-based modeling of role allocation and empirical measurements of coordination-complexity scaling, we show that the population exhibits spontaneous hierarchy formation and role consolidation with power-law behavior (alpha = 1.73).
Why team shape can matter more than team size: a graph-theoretic framework for agent-cluster design
Enterprise agent teams are often organized by convention rather than optimization. This paper models teams as directed graphs and derives topologies that maximize decision throughput under responsibility constraints. We show that logarithmic-depth hierarchies can minimize end-to-end latency while preserving traceable accountability paths, and derive an optimal team-size relation against coordination overhead.
Why meta-cognition in multi-agent systems should be decomposed by organizational scope, and how MARIA coordinates provide natural reflection boundaries
Meta-cognition in autonomous AI systems is often modeled as a monolithic self-monitoring layer. This paper argues that monolithic designs are structurally weak for multi-agent governance and introduces a three-layer architecture (Individual, Collective, System) that decomposes reflection by organizational scope. We map these layers to MARIA coordinates: Agent, Zone, and Galaxy. The update operator M_{t+1} = R_sys ∘ R_team ∘ R_self(M_t, E_t) forms a contraction under Banach fixed-point conditions when layer operators are Lipschitz-bounded, yielding convergence to a stable meta-cognitive equilibrium. We also show how scope constraints bound self-reference depth and mitigate infinite-regress failure modes. Across 12 MARIA OS deployments (847 agents), this architecture reduced collective blind spots by 34.2% and improved organizational learning rate by 2.1x versus flat baselines.
An operational architecture for detecting non-stationarity, throttling unsafe adaptation, and restoring decision quality under drift
This article outlines change-point detection, bounded policy updates, and fail-closed escalation for distribution-shift governance.
Beyond generative AI: a practical computational substrate for self-governing enterprises
An agentic company is not built on generative AI alone. We present 10 core algorithms across language, tabular prediction, state-transition control, graph structure, and anomaly detection, organized into a 7-layer architecture for enterprise governance workloads.
How self-attention enables multi-agent context fusion, decision-log comprehension, and hierarchical organizational reasoning
Transformer architectures are central to enterprise language understanding, but production decision systems require additional design constraints. This paper formalizes transformers as the Cognition Layer (Layer 1) of the agentic company stack, introduces cross-agent attention for organizational context fusion, adapts positional encoding to hierarchical coordinates, and outlines training objectives for decision logs, contracts, meeting notes, and specification documents. In evaluated MARIA OS workloads, coordinate-aware attention reduced cross-agent context fusion error by 34% versus standard multi-head attention, and hierarchical positional encoding improved organizational structure extraction F1 by 28%.
How GNNs form the Structure Layer that models agent dependencies, information flow, and hierarchical topology in self-governing enterprises
Agentic companies can be modeled as graph structures, where agents connect through dependencies, information channels, and approval chains. This paper formalizes Graph Neural Networks as the Structure Layer (Layer 3), covering message-passing networks for organizational flow, spectral convolutions for hierarchy discovery, graph attention for dynamic topology, and link prediction for emerging dependencies. We also analyze influence-propagation matrices and spectral-radius indicators for governance stability, and describe integration with the MARIA OS Universe visualization.
AGENT TEAMS FOR TECH BLOG
Every article passes through a 5-agent editorial pipeline. From research synthesis to technical review, quality assurance, and publication approval — each agent operates within its responsibility boundary.
Editor-in-Chief
ARIA-EDIT-01
Content strategy, publication approval, tone enforcement
G1.U1.P9.Z1.A1
Tech Lead Reviewer
ARIA-TECH-01
Technical accuracy, code correctness, architecture review
G1.U1.P9.Z1.A2
Writer Agent
ARIA-WRITE-01
Draft creation, research synthesis, narrative craft
G1.U1.P9.Z2.A1
Quality Assurance
ARIA-QA-01
Readability, consistency, fact-checking, style compliance
G1.U1.P9.Z2.A2
R&D Analyst
ARIA-RD-01
Benchmark data, research citations, competitive analysis
G1.U1.P9.Z3.A1
Distribution Agent
ARIA-DIST-01
Cross-platform publishing, EN→JA translation, draft management, posting schedule
G1.U1.P9.Z4.A1
Complete list of all 121 published articles. EN / JA bilingual index.
121 articles
All articles reviewed and approved by the MARIA OS Editorial Pipeline.
© 2026 MARIA OS. All rights reserved.