TextMind: Design and Implementation of a Text-Based AI Assistant with Multi-Tier NLP Architecture and Contextual Dialogue Management

The emergence of large language models (LLMs) and advances in natural language processing (NLP) have fundamentally transformed human-computer interaction, making text-based conversational agents a viable medium for delivering intelligent, context-aware information services. This paper presents TextMind, a full-stack text-based AI assistant engineered at Shri Sai College of Engineering and Technology (SSCET), DBATU University, Chandrapur. The system implements a multi-tier NLP architecture comprising a fine-tuned BERT-based intent classifier, Retrieval-Augmented Generation (RAG) with FAISS dense vector search over a domain knowledge base, OpenAI GPT-3.5 Turbo and GPT-4 for open-domain response generation, a finite-state dialogue manager for multi-turn context retention, and a 16-category rule-based NLP fallback layer ensuring 100% non-null response coverage. A browser-accessible Python Flask backend exposes a RESTful chat API consumed by a responsive HTML/CSS/JavaScript frontend with real-time AJAX messaging. Evaluation across 500 test queries achieved intent classification accuracy of 94.6%, entity recognition F1 of 91.3%, BLEU-4 of 0.68, and BERTScore F1 of 0.87. Pilot evaluation with 45 student users yielded satisfaction scores of 4.3–4.5/5.0 across relevance, clarity, and ease-of-use dimensions. TextMind demonstrates that an open-source, lightweight web stack augmented with pre-trained LLM APIs can deliver expert-quality, domain-aware conversational assistance at near-zero infrastructure cost, establishing a reproducible reference architecture for AI assistant development in resource-constrained academic environments.