Uzair Malik

Projects

• BIM-GRASP: A Graph-RAG Approach for Semantic Parsing of IFC Building Models



In this paper, we propose BIM-GRASP, a Graph-RAG-based approach for semantic parsing and natural language querying of IFC building models. The proposed approach integrates IFC data with knowledge graphs and Large Language Models (LLMs) to enable more effective retrieval and interpretation of building information. We evaluate the approach using different building models and query types, demonstrating its effectiveness in accurately retrieving and answering queries related to BIM data.

Available at: (Link)



• LLM-based Content Classification Approach for GitHub Repositories by the README Files



In this paper we have fine tuned several pretrained LLMs to classify different sections of a GitHub README file and our results have outperformed state of the art results. In this paper we have experimented with full fine tuning and parameter efficient fine tuning techniques like Low Ranked Adaptation (LoRA) algorithm.

Available at: (Link)



• Unsupervised Latent Dirichlet Allocation Assisted Github Readme File Content Classification



This project centered around a thesis on machine learning and natural language processing. Its objective was to enhance the quality of GitHub readme files through the utilization of machine learning algorithms. The initial step involved gathering a dataset of GitHub readme files by means of web scraping from the GitHub platform. These files were subsequently annotated manually, adhering to the official recommendations provided by GitHub. Moreover, to automatically label various sections within the readme files, a series of preprocessing steps were implemented. To extract meaningful topics from the readme files and facilitate the automatic labeling process, the latent Dirichlet allocation algorithm, a topic modeling technique, was applied in various configurations. .

Available at: (Link)



• Prediction and Diagnosis of Dengue Fever using Supervised Machine Learning



I oversaw a research project that aimed to develop a diagnostic system for identifying dengue fever in patients based on a range of symptoms. The project involved obtaining a dataset from District Headquarter Hospital Jhelum, which consisted of symptom records from individuals diagnosed with dengue fever. Following a series of preprocessing steps to prepare the data, we employed machine learning algorithms to perform the task of diagnosing and detecting dengue fever in patients.

Available at: (Link)



• A Systematic Literature Review of state of the art algorithms and datasets used for Human Activity Recognition



Performed SLR of state of the art algorithms and datasets used for HAR

Available at: (Link)



• Document Binarization using classic thresholding algorithms and Deep Neural Networks on DIBCO dataset



Conducted a comparative analysis of document binarization approaches with an emphasis on Otsu’s and Niblack’s methods, which belong to the classical approaches of Optical Character Recognition (OCR) and image analysis. Moreover, investigated use of Deep Neural Networks (DNNs) in the given DIBCO database. This study was meant to assess the performance of conventional techniques in thresholding against the state-of-the-art-deep learning in converting gray-scale or color document images to binary forms.



• Pakistan Remote Sensing Satellite Imagery Acquisition System



This is a GIS web based system that allows users to browse and order satellite imagery online.

Available at: (Link)