Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Minds: Empowering Agents with LoRA-as-Tools

Pavan C Shekar, Ashwanth Krishnan

TL;DR

Adaptive Minds addresses the challenge of domain specialization for large language models by introducing LoRA adapters as modular tools and routing queries semantically through the base LLM. The system uses a LangGraph-managed two-agent architecture (Router Agent and Expert Agent) to dynamically select and apply the most relevant domain adapter, enabling on-demand specialization without retraining the backbone model. Key contributions include a fully open-source, end-to-end implementation with 100% routing accuracy on a 25-query test, modest memory overhead (约0.16 GB with five adapters), and a 3.1× mean latency improvement over a non-adapter baseline due to focused, token-efficient generation. The framework demonstrates scalable multi-domain support, adaptability to new adapters, and transparent routing decisions, making it suitable for enterprise assistants, tutoring, healthcare triage, and finance applications while maintaining conversational quality.

Abstract

We present Adaptive Minds, an agentic system that treats LoRA adapters as domain-specific tools. Instead of relying on a single fine-tuned model or rigid rule-based routing, our approach empowers the base LLM itself to act as a semantic router analyzing each query and dynamically selecting the most relevant LoRA tool. This enables the agent to seamlessly switch between different domain experts on demand. By combining the flexibility of multi-agent orchestration with the efficiency of parameter-efficient fine-tuning, Adaptive Minds delivers accurate, specialized responses while preserving conversational ability. The system is built with LangGraph for workflow management, supports both API and web interfaces, and is fully open source, providing a scalable and extensible foundation for domain-adaptive AI assistance.

Adaptive Minds: Empowering Agents with LoRA-as-Tools

TL;DR

Adaptive Minds addresses the challenge of domain specialization for large language models by introducing LoRA adapters as modular tools and routing queries semantically through the base LLM. The system uses a LangGraph-managed two-agent architecture (Router Agent and Expert Agent) to dynamically select and apply the most relevant domain adapter, enabling on-demand specialization without retraining the backbone model. Key contributions include a fully open-source, end-to-end implementation with 100% routing accuracy on a 25-query test, modest memory overhead (约0.16 GB with five adapters), and a 3.1× mean latency improvement over a non-adapter baseline due to focused, token-efficient generation. The framework demonstrates scalable multi-domain support, adaptability to new adapters, and transparent routing decisions, making it suitable for enterprise assistants, tutoring, healthcare triage, and finance applications while maintaining conversational quality.

Abstract

We present Adaptive Minds, an agentic system that treats LoRA adapters as domain-specific tools. Instead of relying on a single fine-tuned model or rigid rule-based routing, our approach empowers the base LLM itself to act as a semantic router analyzing each query and dynamically selecting the most relevant LoRA tool. This enables the agent to seamlessly switch between different domain experts on demand. By combining the flexibility of multi-agent orchestration with the efficiency of parameter-efficient fine-tuning, Adaptive Minds delivers accurate, specialized responses while preserving conversational ability. The system is built with LangGraph for workflow management, supports both API and web interfaces, and is fully open source, providing a scalable and extensible foundation for domain-adaptive AI assistance.
Paper Structure (47 sections, 1 equation, 1 figure, 7 tables)

This paper contains 47 sections, 1 equation, 1 figure, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-Agent Systems for NLP
  4. Parameter-Efficient Fine-Tuning
  5. Domain-Specific Language Models
  6. Intelligent Task Routing
  7. Methodology
  8. System Architecture
  9. Router Agent
  10. Routing Process:
  11. Routing Prompt Template:
  12. Example Prompt:
  13. Expert Agent
  14. Response Generation Process:
  15. Domain Expert Implementation
  16. ...and 32 more sections

Figures (1)

  • Figure 1: Overall architecture of the Adaptive Minds framework. The base LLM acts as a semantic router, dynamically selecting the most relevant LoRA adapter (treated as a tool) to handle each query. The system is flexible—supporting any base model, existing LoRA adapters, and new ones that can be trained and added seamlessly.