Instruction-Tuned Small Language Models for Client Feedback Analysis: A Resource-Conscious Deployment Pipeline

The paper presents a new resource-efficient and privacy-safeguarded Natural Language Processing pipeline which incorporates specialized fine-tuned Small Language Models for distinct operations. Enterprise applications struggle to utilize Large Language Models (LLMs) as general purpose NLP tools because compute costs remain high while implementation in specific domains shows limited success due to privacy requirements. This research addresses LLM limitations by implementing small language models which receive instruction-based fine-tuning in client feedback applications. The integrated system generates organized results that SWOT analysis and text-based trend detection utilize for final assessment. The entire approach consists of domain-specific syntactical tuning and adaptation strategies together with adapter-based mechanism optimization for parameter usage and local computer-based inference and evaluation based on semantic and classical models. The tests show that the adjusted SLM models we evaluated should serve enterprises as a scalable secure NLP subtask solution when compared to LLM technology.