# From LLM to Conversational Agent: A Memory Enhanced Architecture with Fine-Tuning of Large Language Models

**Authors:** Na Liu, Liangyu Chen, Xiaoyu Tian, Wei Zou, Kaijiang Chen, Ming Cui  
**arXiv:** [2401.02777](https://arxiv.org/abs/2401.02777)  
**Date:** 2024-01-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2401.02777v1/

## Key points

- Introduction of RAISE (Reasoning and Acting through Scratchpad and Examples) architecture
- Incorporation of dual-component memory system mirroring human short-term and long-term memory processes
- Enhancement of integration of Large Language Models (LLMs) into conversational agents
- Improving agent controllability and adaptability in complex, multi-turn dialogues
- Phases in comprehensive agent construction scenario: Conversation Selection, Scene Extraction, CoT Completion, Scene Augmentation, LLMs Training
- Preliminary evaluations within real estate sales context showing advantages over traditional conversational agents
- Potential for broader applications beyond real estate domain
- Contribution to AI field by providing a robust framework for developing context-aware conversational agents

## Summary

The paper "From LLM to Conversational Agent: A Memory Enhanced Architecture with Fine-Tuning of Large Language Models" introduces the RAISE (Reasoning and Acting through Scratchpad and Examples) architecture. It aims to enhance the integration of Large Language Models (LLMs) like GPT-4 into conversational agents by incorporating a dual-component memory system that mirrors human short-term and long-term memory processes. This design allows for maintaining context and continuity in conversations, improving agent controllability and adaptability in complex, multi-turn dialogues. The comprehensive agent construction scenario outlined in the paper includes phases such as Conversation Selection, Scene Extraction, CoT Completion, and Scene Augmentation. These steps culminate in the LLMs Training phase, demonstrating RAISE's ability to effectively handle various aspects of conversation processing. Preliminary evaluations conducted within a real estate sales context suggest that RAISE offers advantages over traditional conversational agents. The results indicate its potential for broader applications beyond this specific domain. Authored by Na Liu, Liangyu Chen, Xiaoyu Tian, Wei Zou, Kaijiang Chen, and Ming Cui, this work contributes significantly to the field of AI by providing a robust framework for developing more context-aware and versatile conversational agents. By combining advanced memory systems with large language models and fine-tuning techniques, RAISE represents a promising approach towards creating more intelligent and adaptive conversational agents capable of handling diverse communication scenarios with ease.