Projects

The project proposes an advanced application for infrastructure monitoring that leverages AI and machine learning techniques to address the challenges faced in traditional monitoring systems. The proposed solution aims to automate the monitoring process, identify issues proactively, suggest solutions, and implement them automatically when possible. It involves processing log data, training a GPT model to identify root causes, predict potential failures, and provide recommendations. The application can analyze infrastructure health, job failures, and schedules, offering a comprehensive and proactive approach to monitoring. The solution promises to improve system reliability, accelerate error detection, facilitate faster debugging and resolution, empower engineers, and provide enhanced operational insights. The project includes a proof of concept and a roadmap for further development, incorporating job dependency analysis, predictive dashboards, and alerting mechanisms.

This research focuses on event summarization from news articles utilizing Multi-Document Summarization techniques. Various approaches are implemented and systematically evaluated using metrics tailored to CNN/Daily Mail news articles. The study covers the development and evaluation of both extractive and abstractive text summarization methods. Extractive summarization employs statistical techniques like TF-IDF, TextRank, and Latent Semantic Analysis, while abstractive summarization utilizes neural network models such as pointer-generator seq2seq and transformers. Large pre-trained language models like T5, BART, and LLAMA-2 are fine-tuned on news datasets, and summarization quality is rigorously assessed using metrics like ROUGE and METEOR, along with human evaluation. The research culminates in a dynamic web dashboard facilitating event, topic, and custom article summarization, showcasing automated summarisation capabilities.

The project aimed to address the significant role of fuel prices in shaping the overall inflation landscape, given the crucial implications for economic policy-making and decision-making. By leveraging advanced data science techniques, we sought to enhance the accuracy and reliability of inflation forecasts, particularly in relation to the fuel index.

This project involved creating a data vault system for storing and analyzing medical imaging data obtained from Functional Near Infrared Spectroscopy (fNIRS). The system consisted of different layers, including staging, enterprise, data mart, and GUI layers. The staging layer processed and prepared the data for the subsequent layers, while the enterprise layer stored the data in a warehouse with version control and metadata management. The data mart layer facilitated querying of the stored data, presented in a star schema format. The GUI layer provided a visual interface to display and interpret the brain scan data as images.

This project aimed to predict the successful landing of the Falcon 9 rocket's first stage boosters, allowing SpaceX to offer competitive pricing for satellite launches through booster reuse. Data was collected, normalized, and analyzed using SQL, Python, and visualization techniques. Predictive models were built and evaluated, with the best model used to forecast future landing success.