ENGINEERING BLOG
Technical research and engineering insights from the team building the operating system for responsible AI operations.
121 articles · Published by MARIA OS
Managing student state as high-dimensional vectors with responsibility-gated interventions that prevent harmful over-optimization of learning pathways
Many educational AI systems still optimize around narrow metrics such as test scores, completion rates, or engagement time. Learning, however, is multi-dimensional: knowledge, confidence, motivation, metacognition, and social skills evolve on different trajectories. This paper introduces the Learning State Vector Model, representing each student as a high-dimensional state vector so tutoring agents can make governed decisions across dimensions and reduce harmful single-metric over-optimization.
Why controlling RAG accuracy through responsibility structure outperforms Top-k optimization alone
Many RAG systems optimize retrieval quality primarily through Top-k tuning and embedding similarity. This paper adds a governance-oriented approach: responsibility-tiered gates that adjust validation intensity by risk classification. The framework reports an 82% hallucination-rate reduction on enterprise document corpora while maintaining sub-second response times for low-risk queries.
Responsibility decomposition-point control for enterprise AI agents
When an AI agent modifies production code, calls external APIs, or alters contracts, responsibility boundaries must remain explicit. This paper formalizes fail-closed gates as a core architectural primitive for responsibility decomposition in multi-agent systems. We derive gate configurations via constrained optimization and report that a 30/70 human-agent ratio preserved 97.1% responsibility coverage while reducing decision latency by 58%.
Why investment decisions require conflict management across multiple evaluation universes, not single-score optimization
Traditional investment analysis often compresses multidimensional evaluation into a single score (for example NPV or IRR), which can hide cross-domain conflicts. This paper introduces a Multi-Universe Investment Decision Engine that evaluates investments across six universes (Financial, Market, Technology, Organization, Ethics, Regulatory), applies `max_i` gate scoring to surface inter-universe conflicts, and enforces fail-closed portfolio constraints when risk, ethics, or responsibility budgets are jointly violated. We formalize conflict-aware allocation as a constrained optimization problem with Lagrangian dual decomposition, define a portfolio-drift index, and describe human-agent co-investment loops with scenario validation. Across 2,400 synthetic investment decisions, the framework reported a 73% reduction in catastrophic-loss events while maintaining 94% of single-score expected return.
Formalizing gate strength as a continuous control variable to minimize the combined cost of false positives, missed detections, and investigation delay in AML compliance pipelines
AML programs face a costly tradeoff between false positives, missed detections, and investigation delay. This paper formalizes AML detection as constrained loss minimization over gate strength `g`, derives gate configurations with Lagrangian methods under BSA/FATF/EU AMLD constraints, and reports that MARIA OS responsibility gates reduced total compliance cost by 47% while maintaining required detection thresholds in evaluated settings. The framework treats gate strength as a continuous control variable rather than a binary switch.
How responsibility gates transform multi-agent prisoner's dilemma into a cooperation equilibrium with provable Nash stability
Multi-agent organizations face coordination risk when locally rational behavior converges to defection equilibria. This paper models interactions as iterated prisoner's dilemmas, derives payoff conditions under which defection dominates, and analyzes how responsibility-gate penalties can shift equilibria toward cooperation. We present payoff-matrix construction, gate-penalty design, evidence-forcing mechanisms, and equilibrium-shift analysis. In evaluated settings, cooperation convergence occurred in fewer than 8 rounds with penalty ratios near 0.3x the defection payoff.
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.