Survey on LLM-Powered Chatbots: Architectures, Applications, Challenges, and Future Directions
- Version
- Download 20
- File Size 362.42 KB
- File Count 1
- Create Date 30 November 2025
- Last Updated 30 November 2025
Survey on LLM-Powered Chatbots: Architectures, Applications, Challenges, and Future Directions
Varun Bantia1, Prathyusha K1, Venkatashiva Reddy1, and Dr. Vishwanath Y2 1Department of Computer Science, [Presidency University ,Banglore ], India
2Professor, Department of Computer Science, [Presidency University ,Banglore ], India
Abstract
This survey analyzes nine recent research works in the domain of large language model (LLM)-powered chat- bots, covering their architectures, applications, advan- tages, limitations, and open challenges. Traditional chatbots relied on rule-based or retrieval-based sys- tems, which limited flexibility, adaptability, and scala- bility. With the rise of LLMs such as GPT-3, GPT-4, and LLaMA, conversational agents have become more interactive, context-aware, and capable of performing complex tasks.
The nine studies reviewed in this paper cover do- mains including healthcare, education, nutrition, cy- bersecurity, interdisciplinary research, and sustain- ability. Each paper introduces unique architec- tural innovations—from retrieval-augmented genera- tion (RAG) in clinical decision-making, to multimodal orchestration for interdisciplinary research assistants. Collectively, they demonstrate the versatility of LLM- powered chatbots but also highlight persisting limita- tions such as hallucination, explainability gaps, bias, privacy risks, and unsustainable compute usage.
This survey contributes by providing:
• A taxonomy and classification of the nine works by domain, architecture, and technique.
• A comparative analysis summarizing strengths, limitations, and experimental contexts.
• A synthesis of challenges and open research ques- tions.
• Recommendations for future research directions in chatbot development.
Download