S S

Relational databases remain the primary operational store for enterprise records, but users increasingly expect search behavior that tolerates unfamiliar schemas, synonymy, missing joins, and mixed textual and structured evidence. This paper presents Semantic-Relational Search Integration (SRSI), a framework for adding semantic search and lightweight data integration to existing relational sources without replacing the database engine. SRSI builds a semantic overlay consisting of schema correspondences, attribute phrases, entity links, and query-rewriting templates. Queries are interpreted through lexical, structural, and semantic signals, then compiled into ranked SQL plans and optional ontology-based access paths. The framework is intended for 2020-era deployments: relational systems, SPARQL-to-SQL rewriting, schema matching, record linkage, word and contextual text representations, and conservative human review of low-confidence mappings. A prototype study over product, publication, and service-ticket datasets shows that SRSI improves top-10 answer accuracy from 0.61 to 0.78 relative to keyword-only relational search, reduces manual mapping effort by 42%, and keeps median query latency below 1.8 seconds for the tested workloads. The results suggest that semantic search and relational integration are most practical when semantics are treated as a governed overlay over relational data rather than as a full replacement for relational storage and query processing.

Public cloud providers expose similar high-level resources but differ in processor generations, storage paths, network locality, virtualization overhead, accelerator availability, and pricing rules. These differences make direct comparison difficult for teams that operate analytics, web services, and machine-learning pipelines across providers. This paper presents Provider-Normalized Cross-Cloud Benchmarking (PNCB), a 2021-era framework for measuring and optimizing distributed workloads across heterogeneous public clouds. PNCB separates benchmark definition from provider execution by normalizing workloads into compute, storage, network, and accelerator profiles, then ranking candidate placements by a cost-adjusted service score. The framework combines repeatable workload bundles, synchronized telemetry, policy constraints, and conservative routing rules. A prototype study evaluates batch analytics, key-value serving, containerized microservices, semi-supervised image classification, and GPU stochastic optimization across three public-cloud configurations. Compared with single-provider selection and naive lowest-price routing, PNCB reduces normalized cost by 24-33%, improves p95 completion time by 18%, and reduces placement reversals caused by noisy measurements by 61%. The main result is that cross-cloud optimization requires measurement discipline: provider choice should be based on repeatable workload-specific evidence rather than instance labels or advertised prices alone.

Cloud-native platforms and applied artificial intelligence have matured along partially separate paths. Container orchestration, microservices, declarative configuration, and service-level observability improved software delivery, while production machine learning systems introduced model artifacts, feature pipelines, training jobs, validation gates, and prediction services. Operating these stacks separately creates duplicated control loops, fragile model releases, and limited visibility into the relationship between data drift, model quality, and service reliability. This paper presents Cloud-Native AI Convergence (CNAIC), an architectural pattern that represents AI assets as first-class cloud-native resources and reconciles them through a shared control plane. CNAIC combines model registries, feature freshness checks, serving deployments, drift monitors, and retraining workflows into an auditable release graph. The design is intentionally compatible with Kubernetes-style controllers and 2022-era MLOps practice; it does not assume specialized later foundation-model infrastructure. A controlled study compares script-driven deployment, pipeline-only automation, Kubernetes-only service deployment, and CNAIC across recommendation, document classification, and demand forecasting services. Results show that the converged architecture reduces model release lead time by 39-52%, lowers stale-model exposure by 46-63%, and improves rollback consistency without increasing p95 inference latency beyond 4%. The paper argues that the convergence of cloud-native architecture and AI should be organized around resource reconciliation, provenance, and bounded automation rather than around ad hoc pipeline glue.

Product reviews, support tickets, and survey comments rarely express sentiment as a single document-level label. They mention concrete attributes, such as battery life, delivery, staff, screen, price, or policy, and assign different polarities to each of them. This paper presents Boundary-Aware Attribute Sentiment Extraction (BASE), a joint framework for extracting attribute spans, opinion spans, and attribute-level sentiment labels from review text. BASE combines a pre-trained language encoder with a compact sequence-labeling layer, a span-pair compatibility scorer, and a calibration step that suppresses unsupported sentiment assignments. The framework is designed for settings in which annotated aspect-based sentiment data are limited, domain vocabulary shifts across products, and downstream users need inspectable attribute--opinion evidence rather than only a polarity label. A controlled evaluation on laptop, restaurant, and multi-aspect review benchmarks compares rule-based extraction, CRF tagging, neural sequence labeling, BERT-style sentence-pair classification, and the proposed joint model. Results show that joint span evidence improves attribute sentiment F1 under multi-aspect sentences and cross-domain transfer, while calibrated abstention reduces unsupported sentiment assignments. The study argues that attribute extraction and sentiment analysis should be evaluated as linked evidence recovery rather than as independent extraction and classification subtasks.

Self-ensembling has become a central design pattern for deep semi-supervised learning: a model is trained not only to fit scarce labeled examples, but also to agree with stable targets produced from its own perturbed predictions, historical predictions, or weight-averaged teacher. This paper presents Agreement-Gated Self-Ensembling Review and Refinement (AGSER), a compact framework for analyzing and improving such methods under label scarcity, unlabeled-set shift, and class-dependent confidence imbalance. AGSER separates three decisions that are often coupled in practice: how teacher targets are formed, which unlabeled examples are admitted into the unsupervised loss, and how class-level thresholds are adapted as learning proceeds. The method combines an exponential-moving-average teacher with a two-view agreement gate, class-aware confidence smoothing, and a small audit table that records accepted, rejected, and deferred unlabeled examples. A controlled study on CIFAR-10, SVHN, STL-10, and a shifted unlabeled-pool variant compares temporal ensembling, Mean Teacher, MixMatch, FixMatch-style confidence selection, and adaptive-threshold variants. The results indicate that agreement gates improve stability in the first third of training and reduce high-confidence errors under unlabeled-set shift, while adaptive class thresholds recover examples from weakly learned classes. The paper argues that self-ensembling should be treated as a target-selection and calibration problem, not only as an augmentation-consistency loss.

Peer review depends on independent expert judgment, but high reviewer disagreement can delay editorial decisions, expose inconsistent criteria, and increase the burden on area chairs and editors. This paper presents Reviewer Phrase Agreement Profiling (RPAP), a chronology-scoped framework for estimating when two reviewers are likely to agree before reviews for a target submission are written. RPAP represents each reviewer using prior publications, prior public reviews when available, citation-grounded phrase histories, scientific-document embeddings, and lightweight review-role metadata. The framework explicitly separates reviewer--reviewer affinity from reviewer--submission affinity, blocks information dated after the assignment cutoff, and stores inspectable evidence packs rather than opaque similarity scores alone. A controlled evaluation over open peer-review records compares lexical baselines, citation-informed embeddings, phrase-profile models, and a calibrated RPAP ensemble. The results show that phrase profiles improve agreement-class recall and calibration when paired with reviewer--submission affinity, while open-review data sparsity remains the central limitation. The paper argues that reviewer agreement forecasting should be used as a triage and assignment-audit signal, not as an automated substitute for editorial judgment.

Judicial prediction has progressed from feature-based court-outcome studies to neural and benchmark-driven legal language understanding. Yet many published evaluations still mix outcome identification, retrospective categorization, and prospective forecasting. This paper presents a chronology-scoped framework for historical judicial analysis that separates those settings and treats a prediction as valid only when all features, legal texts, judge histories, and retrieved precedents were available before the target decision. The framework, called Historical Judicial Outcome Modeling (HJOM), combines temporal feature inventory, citation-grounded phrase profiles, long-document encoders, legal-domain pretraining, and calibrated abstention. A controlled analytical study over appellate-style tasks compares lexical baselines, legal transformer encoders, historical profile models, and an HJOM ensemble under random and chronological splits. The results show that chronological evaluation lowers apparent performance but improves interpretability of model failures. The paper argues that useful judicial prediction requires temporal provenance, jurisdiction-aware task definitions, and explanation records that distinguish legal facts, panel history, and data artifacts.

Long-term forecasting has moved from isolated statistical extrapolation toward machine learning pipelines that combine global neural models, exogenous context, decomposition, calibration, and governance controls. This paper reviews those trends from the perspective of 2024 practice and proposes a compact lifecycle framework for selecting and evaluating long-horizon models under chronology, reliability, and operational constraints. The framework, called Chronology-Aware Forecasting Lifecycle (CAFL), separates feature provenance, model-family selection, horizon-specific calibration, evidence retrieval, and drift review. It is motivated by historical-data forecasting, reviewer and legal agreement forecasting, smart-infrastructure analytics, and recent long-sequence time-series models. A controlled analytical study compares four families--recency-weighted regression, recurrent probabilistic forecasting, decomposition transformers, and pretrained time-series models--across stability, calibration, compute cost, and horizon degradation. The results indicate that the most robust long-term forecasting systems in 2024 are not defined by one architecture alone, but by disciplined temporal splits, explicit uncertainty, simple baselines, and auditable evidence around model inputs and forecast claims.

Enterprise knowledge retrieval must search across policy documents, product manuals, incident reports, source repositories, tickets, relational records, and domain-specific analytical stores. Pure lexical retrieval is robust and transparent but can miss paraphrased evidence. Dense retrieval improves semantic recall but may ignore entity scope, temporal validity, permissions, and structured constraints. This paper presents Governed Hybrid Evidence Retrieval (GHER), a retrieval architecture that combines sparse lexical matching, dense passage retrieval, relational filtering, late interaction reranking, and evidence-pack assembly for enterprise knowledge tasks. GHER treats every answerable query as a constrained evidence request: passages are retrieved only within authorized entity scope, scored by multiple retrieval views, fused using calibrated rank aggregation, and returned with structured provenance. A prototype evaluation over policy, support, analytics, and operations corpora shows that GHER improves evidence recall at 10 from 0.64 to 0.78, raises constraint-correct evidence coverage from 0.71 to 0.90, and reduces unsupported response cases by 37.5% compared with dense-only retrieval. The results suggest that enterprise retrieval should be hybrid by design because correctness depends on both semantic relevance and governed structured constraints.

Big data pipelines now combine batch analytics, stream ingestion, machine-learning feature preparation, model-serving telemetry, and governed enterprise data movement. Scale is no longer only a question of adding workers: pipeline operators must select execution engines, partitioning strategies, storage layouts, accelerator routes, and recovery policies while preserving freshness, cost discipline, and compliance. This paper presents Scale-Aware Pipeline Optimization (SAPO), a practical control framework that decomposes a pipeline into typed stages, estimates each stage's bottleneck from telemetry and metadata, and applies bounded optimization actions for placement, partitioning, storage layout, and retry scope. SAPO is designed to operate with established distributed data systems rather than replacing them. A prototype study over batch, streaming, and ML-preparation workloads shows a 31.6% reduction in median completion time, a 27.9% improvement in sustained throughput, and a 22.4% reduction in unit processing cost compared with static configuration, while keeping freshness violations and rollback events within operator-specified limits. The results indicate that scalable pipeline optimization requires coordinated decisions across compute, storage, data quality, and governance boundaries.

Cloud platforms now host continuously changing data pipelines, model-serving systems, API gateways, and compliance-sensitive analytics. Manual governance processes cannot inspect every telemetry stream, configuration change, data movement, and policy exception at the pace of elastic infrastructure. This paper presents Policy-Bounded Agentic Cloud Governance (PACG), a control-plane architecture in which specialized AI agents propose operational actions but cannot execute them without formal policy checks, provenance capture, and staged actuation. PACG separates observation, reasoning, verification, execution, and audit functions so that agentic automation improves responsiveness without weakening accountability. A prototype evaluation across batch analytics, streaming ingestion, and model-serving workflows shows that the approach reduces mean recovery time by 39.8%, reduces unresolved policy drift by 34.6%, and lowers manual intervention events by 62.5% relative to static cloud operations, while keeping p95 actuation latency below 5.4%. The results suggest that the next phase of cloud governance will be neither fully manual nor fully autonomous, but a bounded collaboration among declarative policy, operational telemetry, and auditable agents.

Regulated cyber-physical enterprises increasingly connect foundation-model agents to cloud data pipelines, PMU streams, API gateways, LLM workloads, retrieval systems, calibration programs, anonymized data products, sequential recommendation services, and long-horizon forecasting workflows. The main governance problem is not whether each local system can enforce its own policy, but whether an agent can compose policies across domains before choosing tools, evidence, models, and operational actions. This paper proposes PCFA, a Policy-Composable Foundation Agent architecture. PCFA represents each governed domain as a policy capsule with authority, predicate, privacy, freshness, calibration, risk, tool-contract, and evidence obligations. A composition controller intersects these capsules into a task-specific policy envelope, detects conflicts, routes admissible evidence and tools, and refuses or escalates actions that violate the composed envelope. In a simulated benchmark over 24,000 regulated cyber-physical tasks, PCFA increases policy-valid autonomous completion from 0.68 to 0.91, reduces simulated unauthorized tool attempts from 4.7% to 0.3%, improves audit reconstruction from 0.73 to 0.96, and reduces adverse-action CVaR by 39.1% relative to the strongest contract-only baseline.

Enterprise AI agents need representative records for retrieval testing, policy evaluation, incident rehearsal, risk analysis, and documentation maintenance, but using raw operational data in these workflows can expose sensitive fields and silently reinforce stale policy assumptions. This paper proposes RSDG, a reflexive synthetic data governance architecture for enterprise AI agents. RSDG treats synthetic-data generation, anonymization, retrieval, validation, documentation, release, and downstream agent feedback as one governed lifecycle. A release candidate is accepted only when it satisfies privacy accounting, hybrid evidence support, model-context contract validity, risk constraints, and phrase-level policy-drift checks. The system also observes how released synthetic records affect agent answers and decisions, then renews, constrains, or withdraws datasets whose utility, privacy risk, or narrative influence has changed. In a simulated benchmark spanning compliance evidence, policy question answering, and financial risk review, RSDG improves evidence-supported synthetic-record acceptance from 0.72 to 0.92, reduces simulated disclosure-risk alerts from 8.4% to 1.1%, and cuts unsupported policy-influence events by 63.5% relative to an unbounded synthetic-generation baseline.

Cross-cloud artificial intelligence platforms increasingly serve both flexible large-language-model (LLM) workflows and time-critical power-grid analytics. Carbon-aware scheduling can shift model work toward lower-emission regions and times, but indiscriminate shifting can delay phasor measurement unit (PMU) event detection, lose calibration warnings, violate data-location policy, or make model outputs unavailable when operators need them. This paper proposes CACS, a Carbon-Aware Cross-Cloud Scheduler for LLM and grid analytics. CACS fuses forecast carbon intensity, cloud capacity and latency telemetry, governed API intents, PMU digital-twin state, calibration risk, and job-specific service guarantees. It admits or routes strict grid jobs immediately through feasible low-latency paths while spatially or temporally shifting deferrable LLM inference, retrieval refresh, and replay analysis under explicit deadline and policy constraints. A simulated 30-day benchmark with three cloud regions, two grid edge zones, and mixed PMU and LLM traffic shows that CACS reduces estimated operational emissions by 31.6% compared with latency-first approved-cloud routing while retaining a 242 ms P95 urgent-grid decision latency, 98.8% deadline satisfaction, and zero simulated policy violations.

Legal, compliance, and enterprise-policy retrieval systems must explain not only which source supports an answer, but also which recurring phrase, standard, exception, or interpretive frame shaped the answer. Conventional retrieval-augmented generation can cite documents while still missing the lineage of policy language across runbooks, legal memoranda, contracts, incident reports, and governance repositories. This paper proposes Meme-Aware Legal and Policy RAG (MAP-RAG), a synthetic architecture that tracks phrase-level policy memes as auditable units in a retrieval graph. MAP-RAG combines hybrid semantic-relational retrieval, critic-guided documentation maintenance, distributed RAG, contract-mediated agent tools, privacy-preserving evidence views, and drift-aware phrase scoring. It extends Retrieval-Grounded Documentation Agents for Enterprise Compliance Evidence with phrase-level explanation packs, extends Ideological Drift Detection in Governed Enterprise Knowledge Bases with legal and policy influence graphs, and extends Contract-Driven Multi-Agent Incident Response for Cloud-Native Platforms with governance-specific tool contracts. In a simulated legal and enterprise-policy benchmark, MAP-RAG improves citation-supported answer faithfulness from 0.74 to 0.91, raises phrase-lineage explanation coverage from 0.38 to 0.86, and reduces unsupported policy actions by 52.8% relative to a strong RAG-only baseline.

Regulated personalization systems must decide what information, offer, limit, explanation, or next action to present while preserving privacy and controlling downstream harm. Conventional sequential recommenders optimize engagement over behavior traces, but regulated settings also require privacy budgets, evidence constraints, risk limits, and auditable tool use. This paper proposes Privacy-Preserving Sequential Decision Systems (PPSDS), a synthetic architecture for regulated personalization that combines efficient sequence modeling, differentially private behavior summaries, constrained distributional policy gradients, hybrid evidence retrieval, distributed RAG, and contract-bound agent interoperability. PPSDS extends Ideological Drift Detection in Governed Enterprise Knowledge Bases with a drift sentinel for personalization narratives and extends Risk-Aware Financial RAG with Distributional Retrieval Policies with action-level privacy and tail-risk gates. In a simulated personalization benchmark spanning finance, benefits, and customer-support recommendations, PPSDS preserves most of the utility of an unconstrained sequential policy while reducing simulated privacy exposure by 72.0%, lowering conditional value-at-risk of adverse outcomes by 41.5%, and eliminating contract-violating evidence accesses.

Cloud-native incident response increasingly spans Kubernetes clusters, API gateways, vector stores, observability backends, model endpoints, and compliance repositories. Human responders need fast evidence, but unbounded automation can make incidents worse by invoking unsafe tools, leaking sensitive context, or applying stale runbook steps. This paper proposes Contract-Driven Multi-Agent Incident Response (CD-MAIR), a synthetic architecture in which specialized response agents negotiate typed tool contracts before they inspect systems, retrieve evidence, update runbooks, or shift workloads across cloud providers. CD-MAIR combines governed pipeline agents, secure API-gateway intents, critic-guided documentation maintenance, distributed retrieval-augmented generation, model-context-protocol contracts, cross-cloud workload placement, and privacy-preserving evidence views. It extends Policy-Verified Agentic DataOps for Regulated Multi-Cloud Analytics with incident-specific action verification, extends Retrieval-Grounded Documentation Agents for Enterprise Compliance Evidence with live runbook repair, and extends Cross-Cloud LLMOps Scheduler for Privacy-Budgeted RAG and Inference with incident-aware route selection. In a simulated benchmark over cloud-native outages, CD-MAIR reduces mean time to qualified mitigation from 41.8 to 24.6 minutes, improves evidence-supported action rate from 0.73 to 0.92, and eliminates unauthorized tool invocations under the benchmark policy.

Enterprise retrieval-augmented generation (RAG) and large-language-model inference increasingly run across multiple cloud providers, vector stores, data catalogs, and model endpoints. The operational scheduler must therefore optimize latency and cost without violating data residency, gateway intent policies, model-context-protocol contracts, or privacy budgets. This paper proposes a Cross-Cloud LLMOps Scheduler (CCLS), a synthetic architecture for routing extraction, retrieval, context construction, and inference under explicit privacy accounting and evidence-maintenance constraints. CCLS extends Policy-Verified Agentic DataOps for Regulated Multi-Cloud Analytics by applying policy-verified execution to LLM workloads, and it extends Retrieval-Grounded Documentation Agents for Enterprise Compliance Evidence by treating compliance evidence freshness as a first-class scheduling signal. The scheduler combines governed API intents, cross-cloud workload placement, distributed RAG, MCP tool contracts, anonymized evidence views, and latency-aware sequence models for context selection. We define the architecture, a multi-objective scheduling model, and a simulated benchmark over compliance Q&A, operational summarization, and governed decision-support traffic. In simulation, CCLS reduces P95 request latency by 40.5% relative to static approved-cloud routing, improves evidence-supported answer recall from 82.4% to 91.8%, and eliminates privacy-budget overruns and unauthorized policy violations.

Our aim Is to give an idea of the Nyaya and Advaita tlieories of perception and to note metapliysicai or ontological elements In them. We shall consider whether it is possible to sieve out features of the theories without such elements with a view to formulating a commonly acceptable platform for dialogue regarding a theory of perception. In recent times scholars have attempted to pick up common elements in the two theories. In our account we may, however, be allowed to use Sanskrit philosophical words in original. This is perhaps useful for philosophizing freely in a comparative setting.

Our aim is to give an idea of the Nyaya and Advaita theories of perception and to note metaphysical or ontological elements in them. We shall consider whether it is possible to sieve out features of the theories without such elements with a view to formulating a commonly acceptable platform for dialogue regarding a theory of perception. In recent times scholars have attempted to pick up common elements in the two theories. In our account we may, however, be allowed to use Sanskrit philosophical words in original. This is perhaps useful for philosophizing freely in a comparative setting.

Focusing on the philosophical puzzle of time and its relation with being and presence the paper explores the volatile relationalities un/tying them in shaping our conceptualisation of memory as re-turning. With such an approach the paper analyses the paradoxes that always haunt any attempt at thinking time, being and presence in their specificity as well as within their general embrace. It is through such play of the specific and general, the paper submits, that the thinking of memory and its acts of re-turning comes to be conceptualised in terms of linear, teleological, homogenous understanding of continuity. Turning towards Bergson and Heidegger’s approaches to the question of time, presence and being the paper attempts to open-up the layered paradoxes that not only shape any act of turning but also the thinking of possible itself as a general category for conceptualising memory as re-turning. Bringing in the question of language and its ontological and temporal concerns, the paper thus brings in the concept of “poetics” to hint at the continuous negotiation that thinking of such interstices in language demands. It is towards such contingencies of the (un)timely, the paper submits, that any attempt at thinking memory as re-turning gestures.