Christos Smarlamakis
PhD Candidate, Department of Business and Organizations Administration.
University of Peloponnese
Efstratios Georgopoulos
Professor of Artificial Intelligence, Department of Business and Organizations Administration.
University of Peloponnese
In the contemporary era of exponential knowledge generation, the phenomenon of severe Information Overload poses a critical and escalating bottleneck for corporate Research and Development (R&D) departments, academic institutions and Business Intelligence (BI) ecosystems. The rapidly accelerating pace of scientific discovery, coupled with the unprecedented proliferation of digital archives and unstructured data repositories, has fundamentally altered the landscape of competitive intelligence. Traditional methodologies for systematic literature reviews and market analysis are heavily reliant on manual intervention. Consequently, these established protocols are prohibitively resource-intensive, difficult to scale, and highly susceptible to inherent human cognitive biases. While the advent of Generative Artificial Intelligence and Large Language Models (LLMs) has introduced promising avenues for automated text analysis, the prevailing reliance on commercial, cloud-based AI solutions introduces significant strategic vulnerabilities. For enterprise-scale applications, processing thousands of complex documents via external commercial APIs incurs unsustainable operational expenditure (OPEX) due to token-based pricing models. More importantly, transmitting sensitive corporate strategies, proprietary R&D inquiries, or classified technological research to third-party cloud servers poses severe data privacy risks, potentially violating regulatory compliance frameworks (such as the GDPR) and exposing organizations to corporate espionage.
To address these intertwined technical, economic, and security challenges this paper presents TALOS (Tactical Agentic Literature Orchestration System), a novel, autonomous Decision Support System (DSS) designed specifically for cost-aware, multi-provider bibliographic orchestration and advanced Business Intelligence. TALOS fundamentally reimagines the systematic Knowledge extraction pipeline by operating as an independent, modular AI agent that autonomously executes the core, labor-intensive phases of established research protocols, such as the PRISMA framework (Identification, Screening, Eligibility, and Inclusion). By shifting the paradigm from passive Retrieval-Augmented Generation (RAG) to an active, “agentic” workflow, the proposed framework significantly enhances the efficiency of competitive intelligence gathering across a vast and heterogeneous landscape of academic and commercial data sources.
At its core, TALOS decouples the mechanical process of data ingestion from the higher-level task of cognitive evaluation. The system incorporates an adaptive, cost-aware machine learning agent that functions as an intelligent “API Forager”. Instead of relying on static, hard-coded heuristics, this agent continuously learns an optimal policy for navigating a complex, high-dimensional state space. This state representations encompasses real-time source usage metrics, historical error patterns, and dynamic provider-level rate limits. Consequently, the agent can make informed, sequential decisions regarding which of the integrated academic repositories or business databases to query next, maximizing data retrieval while strictly minimizing latency and server blocks. Once the raw data is ingested into a secure, ACID-compliant local database hub, TALOS applies a proprietary Quad-Layer cognitive evaluation framework. This mechanism provides a structured, multi-dimensional lens to autonomously assess the relevance and quality of the literature across four distinct, user-defined axes: Strategic planning, Operational mechanisms, Tactical Implementation, and Applied simulation environments.
Designed for maximum operational resilience and strict cost control, the architecture of TALOS is fortified by an AI Manager module featuring a custom-built dynamic fallback routing and circuit-breaker mechanism. This orchestrator is capable of dispatching requests across multiple LLM providers, strategically prioritizing free tiers and open-source models to aggressively minimize operational expenditure. During multi-provider evaluations, any external API failure, rate limit breach (e.g. HTTP 429 errors), or token exhaustion automatically triggers the Circuit Breaker. Upon exceeding a configurable failure threshold, the workflow seamlessly and instantaneously reroutes all subsequent cognitive evaluation tasks from expensive cloud endpoints to locally hosted, air-gapped Edge LLMs. This hybrid, self-healing architecture enables fully offline operation, ensuring uninterrupted data analysis, zero-downtime performance, and absolute data sovereignty even under complete external network failure.
The proposed framework’s efficacy, scalability, and economic performance were empirically validated through a comprehensive, multi-layered systematic review targeting a highly complex, real-world engineering problem: Cooperative Mission Planning for Autonomous Robot Swarms. During the execution of this study, the autonomous agent successfully navigated multiple external databases, processing thousands of raw bibliographic records. Utilizing a multi-provider hybrid embeddings system, TALOS generated high-dimensional semantic vectors, facilitating advanced semantic search capabilities to ensure consistent and highly accurate vector comparisons. Ultimately, the system demonstrated significant improvements in resource efficiency, autonomously isolating a highly selective subset of “Elite”, top-tier publications without requiring manual human oversight. Deployed as a 24/7 autonomous research service equipped with multi-channel notifications and a lightweight API for real-time monitoring, TALOS provides a domain-agnostic and highly scalable blueprint for automating knowledge extraction. The architecture presented in this study holds profound strategic implications for highly regulated, data-intensive sectors such as financial analysis, legal tech research, and corporate business intelligence, where data integrity, OPEX minimization, and operation resilience are of paramount importance.
Keywords: Decision Support Systems, Autonomous Agents, Business Intelligence, Generative AI, Cost-Effective.
JEL Classification Codes:
- O32 (Management of Technological Innovation and R&D)
- C61 (Optimization Techniques – Programming Models – Dynamic Analysis)
- M15 (IT Management)