Faheem Uddin Tayyab Mohammed

Client: Duradiamond, Perth

Project: Duradiamond Healthcare – Patient Portal

Project Description: Duradiamond Patient Access online appointment bookings and cancellations means less time spent on hold. Ensure patients get their regular medication when they need it with online repeat prescription ordering and automated delivery to their preferred pharmacy. Patients can view test results online, saving them from having to travel to the practice for an unnecessary appointment. Availability 24-hours a day means that patients don’t have to wait for practice opening times and can access services whenever they need. Patients can securely share their medical record – an important feature if they need care abroad, out of hours, or in an emergency situation. Shared GP record history allows patients to view an audit of where, when and why someone has accessed their record using a data sharing agreement.

Website: https://www.duradiamondhealth.com

Responsibilities:

• Led end-to-end design and implementation of predictive models to optimize patient appointment bookings, repeat prescription forecasts, and test result delivery timelines using supervised learning techniques (Random Forest, XGBoost, Logistic Regression), improving service availability by 34%.
• Performed advanced exploratory data analysis (EDA) and statistical modeling using Python (NumPy, Pandas, Seaborn, Scikit-learn) to identify patient behavior patterns, sentiment indicators, and cancellation risks, enhancing the personalization engine across the portal.
• Developed and deployed machine learning pipelines on Azure Databricks and Azure Data Lake to support real-time recommendation systems and anomaly detection for secure medical record access, ensuring compliance with GDPR and HIPAA protocols.
• Built and visualized patient engagement dashboards using Power BI and Excel, integrating insights from SQL Server and cloud storage layers to present actionable KPIs to stakeholders, and deployed deep learning models (CNNs, RNNs using TensorFlow & Keras) for OCR-based medical data classification.

Environment: Python, R, SQL, Pandas, NumPy, Scikit-learn, TensorFlow, Keras, Matplotlib, Seaborn, Azure Data Lake, Azure Databricks, AWS Sagemaker, GCP, Power BI, Tableau, Excel, SQL Server, Oracle, Jupyter Notebook, Anaconda, GitHub, TensorBoard.

Client: NHS

Project: NHS AI Assistant – Dr.GPT Bot

Project Description: The Dr.GPT Bot project aimed to develop an AI-powered assistant for the NHS TaskLearn Application, capable of answering quiz-based questions using a hybrid Retrieval-Augmented Generation (RAG) approach. Data was sourced from the TaskLearn App's MongoDB (ConvertedQuizzes collection) and ingested into Azure Data Lake Storage using Azure Data Factory. A pipeline was built to automate data extraction, transformation, and loading. The model architecture utilized FAISS for semantic search and a fallback mechanism invoking a fine-tuned Google FLAN model via Hugging Face Transformers. A Streamlit-based user interface was created to allow natural language interactions with the bot. The final solution provided reliable, context-aware responses, improving accessibility and engagement with quiz data for NHS users.

Website: http://65.0.204.147/ and http://65.0.204.147/api/

Responsibilities:

• Designed and deployed an AI-powered chatbot (Dr.GPT) using a hybrid RAG (Retrieval-Augmented Generation) architecture, integrating FAISS-based semantic search and a fine-tuned Google FLAN transformer model via Hugging Face to enable accurate quiz question answering from NHS TaskLearn data.
• Built robust ETL pipelines using Azure Data Factory to ingest quiz content from MongoDB into Azure Data Lake Storage, automating transformation workflows, ensuring scalable ingestion, and enabling real-time data access for downstream ML processes.
• Implemented data preprocessing, EDA, and feature engineering using Python libraries (Pandas, NumPy, Scikit-learn) to clean, normalize, and encode question-answer pairs for model readiness, including exploratory analytics to assess text distributions, sentiment patterns, and keyword density.
• Developed, fine-tuned, and evaluated transformer-based models for natural language understanding, applying techniques such as tokenization, cosine similarity, embeddings optimization, and fallback handling with confidence thresholds to ensure context-aware, reliable bot interactions.
• Integrated the Dr.GPT solution into a Streamlit-based front-end UI, allowing NHS staff to interact with the model via natural language queries; supported secure API deployment, versioning via GitHub, and continuous integration pipelines using Azure DevOps.

Environment: Python, R, SQL, Pandas, NumPy, Scikit-learn, FAISS, Google FLAN, Hugging Face Transformers, Azure Data Factory, Azure Data Lake Storage, MongoDB, Streamlit, GitHub, Power BI, TensorFlow, Keras, Jupyter Notebook, Anaconda, Azure DevOps.