SPARTAN AI Master Companion: An Intelligent Orchestration System for Multi-Domain AI Assistants

Abstract

Modern professionals face complex tasks that span multiple domains of expertise. While AI assistants such as ChatGPT have become commonplace, users must either rely on generalist models with limited domain expertise or manually coordinate multiple specialized tools, a time-consuming process that imposes significant cognitive overhead. SPARTAN AI’s Master Companion addresses this by introducing an intelligent orchestration system that automatically identifies, selects, and coordinates specialized AI assistants to handle multi-domain requests seamlessly. The system leverages Microsoft Phi-3 Mini (3.8B parameters) with a hybrid architecture combining keyword-based classification and LLM reasoning for intelligent routing. It employs advanced orchestration algorithms, including sequential execution with context propagation, quality validation with intelligent retry mechanisms, graceful degradation for service failures, and real-time progress streaming via Server-Sent Events. The Master Companion orchestrates 11 specialized assistants across diverse domains, including fitness coaching, software development, content creation, and financial planning. Built using Next.js 16, React 19, Python FastAPI, Convex real-time database, and Docker, the system provides dynamic model selection across seven frontier models (Claude Opus 4, GPT-5, OpenAI O1, Gemini 2.0 Flash, Qwen 3 235B, Grok 4.1 Fast, DeepSeek V3). Production deployment demonstrates routing precision > 0.90, recall > 0.85, and workflow completion under 40s while maintaining explainability and user trust through human-in-the-loop approval.

Introduction

Recent advances in large language models (LLMs) have enabled powerful AI assistants, but users face a trade-off between generalist models (broad but shallow knowledge) and coordinating multiple specialized tools manually. Multi-domain tasks, like combining fitness advice with email drafting, require orchestrating multiple domain-specific assistants efficiently.

The paper presents SPARTAN’s Master Companion, a production-ready multi-agent orchestration system that:

1. Automatically selects relevant assistants.
2. Determines execution order and dependencies.
3. Propagates context across assistants for multi-stage workflows.
4. Handles failures gracefully.
5. Synthesizes outputs into coherent responses.

Key contributions include:

• Hybrid Routing Architecture: Combines sub-10ms keyword-based classification with Phi-3 Mini LLM reasoning for explainable assistant selection, execution order, and fallback handling.
• Context Propagation System: Sequential execution engine ensures outputs from earlier assistants inform later stages.
• Empirical Validation: Achieves >0.90 routing precision, >0.85 recall, >0.95 success rate, with workflows under 40s.

The system implements 11 domain-specialized assistants for tasks including fitness coaching, email writing, software development, grammar correction, and academic tutoring. Execution relies on a dual-stage routing: fast keyword filtering followed by LLM planning. The workflow engine manages context, validates output quality, and ensures graceful degradation if components fail, producing coherent multi-assistant results.

Conclusion

SPARTAN AI’s Master Companion demonstrates that coordinated specialized AI assistants can outperform single generalist models while maintaining production reliability. The system combines hybrid routing (keyword + Phi-3 Mini LLM), comprehensive graceful degradation, and real-time SSE streaming to orchestrate 11 specialized assistants in diverse domains. Key contributions include: 1) Explainable routing with LLM-generated reasoning and confidence scores 2) Sequential execution with context propagation enabling multi-stage workflows 3) Fallback strategies ensuring availability under service failures 4) Transparent workflow progress during 30–45s executions Production deployment achieved routing precision > 0.90, recall > 0.85, and success rate > 0.95 while completing workflows in under 40s. Critical insights from deployment include the necessity of graceful degradation for AI service reliability, the importance of explainability for user trust, and the value of real-time feedback for perceived responsiveness. Future work will target faster routing, reinforcement learning from feedback, and distributed parallel execution.

References

[1] OpenAI, “GPT-4 Technical Report,” arXiv preprint arXiv:2303.08774, 2023. [2] Anthropic, “Introducing Claude,” Anthropic Blog, 2023. [3] M. Wooldridge and N. R. Jennings, “Intelligent Agents: Theory and Practice,” The Knowledge Engineering Review, vol. 10, no. 2, pp. 115–152, 1995. [4] Significant Gravitas, “AutoGPT: An Autonomous GPT-4 Experiment,” GitHub, 2023. [5] H. Chase, “LangChain,” GitHub, 2022. [6] Q. Wu et al., “AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation,” arXiv:2308.08155, 2023. [7] S. Hong et al., “MetaGPT: Meta Programming for Multi-Agent Collaborative Framework,” arXiv:2308.00352, 2023. [8] I. Ong et al., “RouteLLM: Learning to Route LLMs with Preference Data,” arXiv:2406.18665, 2024. [9] T. Shnitzer et al., “Large Language Model Routing with Benchmark Datasets,” arXiv:2309.15789, 2023. [10] “LLM Benchmarks - Model Performance Rankings,” LLM-Stats, March 2026. [Online]. Available: https://llm-stats.com/benchmarks

Copyright

Copyright © 2026 Sunil Shakhapure, Rohit Shalgar, Uday More, Bhargav Katkam, Anuja Narsale. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.