Journal of Intelligent Decision Technologies and Applications (e-ISSN: 3049-0219)

AI-Based Travel Planning System Using Large Language Models

Vedant Desai — 2026-02-07

Travel planning is a multi-constraint problem that involves balancing budget, duration, transportation, accommodation, and personal preferences. Existing online travel platforms largely focus on booking individual services and often fail to generate cohesive, personalized itineraries that strictly adhere to user-defined financial limits. This paper presents the design and implementation of an AI-based travel planning system that generates realistic, budget-constrained travel itineraries using Large Language Models (LLMs). The proposed system adopts a modular three-layer architecture consisting of a web-based frontend, a backend orchestration layer, and an AI inference layer hosted on Groq. User inputs such as destination, total budget, number of days, and travel preferences are transformed into structured prompts using strict prompt engineering techniques. Unlike conventional approaches where budget is treated as contextual information, this system enforces budget as a hard constraint, ensuring that the generated itineraries do not exceed the specified financial limit. The system outputs a structured itinerary comprising a budget breakdown, day-wise travel plan, and practical local tips, formatted in controlled HTML for consistent presentation. Additionally, transport booking options are abstracted at the user interface level to avoid dependency on region-specific service providers, enhancing scalability and geographic neutrality. Experimental observations indicate that explicit prompt constraints significantly improve itinerary realism, cost adherence, and output consistency. The results demonstrate that careful prompt design combined with modular system architecture enables Large Language Models to function effectively in real-world, constraint-driven applications such as travel planning.

An Integrated Approach of the Modern Expert System and Its Application

Padma Lochan Pradhan — 2026-02-28

This research paper examines how modern integrated expert systems democratize decision-making by simulating specialized judgment in complex, domain-specific problems. The central thesis is that these systems provide consistent and efficient access to expertise for non-experts. To demonstrate this, the authors design solutions focused on knowledge preservation, standardized decision-making, rapid problem-solving, and ensuring justification and transparency. A significant finding in 2024 is the shift towards hybrid AI models that combine structured knowledge (rules) with machine learning capabilities, moving away from older standalone rule-based systems. This approach has shown improved performance, with hybrid methods achieving better accuracy in complex diagnostic scenarios compared to single-method systems. By late 2024, a large percentage of expert systems had incorporated machine learning. Knowledge Preservation: Capturing the specialized knowledge of human experts and storing it in a digital format ensures that valuable expertise is not lost when professionals retire or leave an organization. Ultimately, the sustainability of fuzzy logic technology leads to improved quality, performance, better cost management, enhanced decision-making, and minimized risk. By incorporating insights from academic studies, industry analyses, and practical applications, this paper provides a comprehensive overview of the evolution and innovation of modern expert systems. Expert system performance matrices focus on accuracy, speed, and domain-specific metrics (like sensitivity/specificity in medicine), evaluated using ROC curves, AUC, precision, recall, F1-score, and validation methods like cross-validation, with applications spanning medical diagnosis (prostate cancer), process control, and fault diagnosis, aiming to match or exceed human expert capabilities for complex decision-making. Current research focuses on automating complex decision-making in domains where human expertise is scarce or expensive, such as medical diagnostics, industrial maintenance, and personalized education.

Implementation of an AI-Assisted Textile Waste Valorization Platform for Circular Fashion Ecosystems

Kanksha R. K — 2026-02-28

The textile and apparel industry generates a significant amount of pre-consumer fabric waste during manufacturing processes such as cutting, stitching, and finishing. Although much of this waste retains functional and aesthetic value, it is often discarded due to the lack of structured reuse mechanisms. At the same time, artisan communities engaged in craft-based and sustainable fashion practices face ongoing challenges in accessing affordable and consistent raw materials. This gap between waste generation and material demand highlights the need for effective digital solutions that support reuse and circular fashion. This paper presents PunarVastra, an AI-assisted textile waste valorization platform designed to connect textile factories with artisans through a lightweight and accessible digital ecosystem. The platform allows factories to upload images and basic details of fabric scraps, which are processed using an AI-assisted analysis module to generate standardized descriptors such as color and texture. These classified materials are then made available through a digital marketplace, enabling artisans to easily browse, evaluate, and reuse suitable textiles for their craft and production needs. The system is implemented using a web-based frontend, a Flask-based backend, and a simulation-driven AI module to ensure compatibility with resource-constrained environments. Experimental evaluation demonstrates that the platform effectively supports digital documentation, classification, and retrieval of textile waste with minimal computational overhead. The proposed solution highlights the potential of combining digital platforms and AI-assisted processes to reduce textile waste, improve material accessibility, and promote sustainable and circular fashion practices.

Human-Centered Context-Aware Student Stress Detection Using LSTM- Based Academic Behavior Analytics

S. Omprakash — 2026-03-30

Academic stress has become a major concern among students in modern educational environments. Increasing academic pressure, frequent examinations, tight assignment deadlines, and the extensive use of digital learning platforms contribute significantly to students’ mental stress. If not identified at an early stage, prolonged stress can negatively affect academic performance, emotional well-being, and overall student development. Therefore, early detection of stress is essential for providing timely academic and psychological support. This study proposes a human-centered and implementation-oriented framework for detecting student stress using sentiment analysis and deep learning techniques. The proposed system utilizes a Long Short-Term Memory (LSTM) neural network to analyze textual feedback collected from students through surveys, discussion forums, and online learning platforms. The collected text data undergo several preprocessing steps, including cleaning, tokenization, stopword removal, and sequence padding, to improve data quality and model learning capability. The LSTM model is designed to capture contextual and sequential patterns present in student expressions, enabling accurate classification of text into stress and non-stress categories. Experimental evaluation demonstrates that the proposed approach achieves reliable performance in identifying stress-related sentiments from student feedback. The system provides a practical tool for educational institutions to monitor student well-being and implement early interventions. Overall, the proposed method highlights the potential of deep learning and sentiment analysis in supporting student mental health within academic environments.

Smart Cost Estimation for Construction Planning: An Automated Framework Leveraging Computer Vision and Collaborative Filtering

Snehal Ghoparkar — 2026-04-13

In the traditional construction landscape, particularly within developing economies, the pre-planning phase is severely bottlenecked by manual quantity take-offs (QTO) from two-dimensional blueprints. This conventional approach is labour-intensive, computationally opaque, and susceptible to human error, often leading to budget variances exceeding 20%. This paper introduces a novel “Smart Hybrid” framework designed to democratise professional-grade estimation for Small and Medium Enterprises (SMEs). Moving beyond fragile edge-detection techniques, the solution integrates a multi-stage Hybrid Vision Pipeline. This pipeline uniquely synergises the semantic reasoning of large language models (specifically Google Gemini 2.0) to interpret room topology, with the geometric precision of OpenCV and PaddleOCR for exact dimension extraction. Furthermore, the system incorporates a deterministic civil engineering logic engine that strictly adheres to IS 1200 standards. For visualisation, a Generative Layout algorithm using Breadth-First Search (BFS) to reconstruct 2D plans into interactive 3D models via React Three Fibre. Finally, procurement is optimised through a recommender engine applying collaborative filtering to real-time supplier data from IndiaMART.