Anam Rafique

Led the business analysis for PrivetLM, a custom enterprise language model deployed on Azure.
Conducted stakeholder meetings and requirement gathering to define use cases, success metrics, and deployment goals.
Authored and maintained comprehensive and technical documentation including Functional Requirements Document (FRD), Use Case Specifications, User Manuals, Website Content describing PrivetLM’s capabilities and integration, Meeting Minutes and Weekly Summaries to track progress and decisions.
Facilitated cross-functional collaboration between business teams and AI engineers to ensure clarity, feasibility, and timely delivery.
Collaborated with team for Retrieval-Augmented Generation (RAG) architecture with the Microsoft Azure AI stack.
Developed POCs for different use cases using Generative AI and libraries like Langchain.
Collaborated with a cross-functional team (AI/ML, DevOps, front-end/back-end) and USA leadership to support demos, and competitive analysis against market competitive tools.
Designed and developed a conversational agent using Microsoft Copilot Studio, leveraging a public website as the primary knowledge source.
Structured and introduced diverse topics into the agent’s knowledge base to ensure comprehensive and context-aware responses.
Configured trigger phrases, topics, and fallback mechanisms to enhance user interaction and coverage.
Integrated the Copilot agent with Microsoft Teams, enabling seamless access to website-based knowledge within the enterprise communication environment.
Conducted iterative testing and refinement of the agent’s behavior to improve accuracy, relevance, and user experience.
Designed and implemented a modular text and image-generation pipeline in Jupyter Notebook integrating multiple generative models and external tools to produce high-quality synthetic images.
Developed preprocessing and postprocessing routines (image normalization, augmentation, resizing, filtering) using PIL/OpenCV and NumPy to improve model input quality and output consistency.
Integrated Hugging Face model APIs and Diffusers pipelines to run text-to-text, text-to-image and image-to-image generation, optimizing prompts and scheduler settings for quality and inference speed.
Led the Customer Segmentation use case using K-Means and Hierarchical Clustering to identify customer groups based on service usage, tenure, and billing patterns.
Engineered features from raw Telco data, including derived metrics like monthly usage trends and contract stability indicators.
Applied SMOTE to balance the dataset and improve model fairness across churn classes.
Conducted Exploratory Data Analysis (EDA) to uncover churn drivers and visualize customer behavior trends.
Collaborated on Churn Prediction using Logistic Regression, Random Forest, and Gradient Boosting, achieving high ROC-AUC scores.
Supported Service Recommendation using Collaborative Filtering and Content-Based Filtering to personalize offers.
Deployed models via FastAPI, enabling real-time inference for churn risk and service recommendations.
Integrated Power BI Dashboards to visualize churn risk heatmaps, segmentation clusters, and recommendation insights.
Authored Functional Requirement Documents (FRDs) and Solution Design Documents detailing architecture, workflows, and cost estimates.

Led the business analysis for PrivetLM, a custom enterprise language model deployed on Azure., Conducted stakeholder meetings and requirement gathering to define use cases, success metrics, and deployment goals., Authored and maintained comprehensive and technical documentation including Functional Requirements Document (FRD), Use Case Specifications, User Manuals, Website Content describing PrivetLM’s capabilities and integration, Meeting Minutes and Weekly Summaries to track progress and decisions., Facilitated cross-functional collaboration between business teams and AI engineers to ensure clarity, feasibility, and timely delivery., Collaborated with team for Retrieval-Augmented Generation (RAG) architecture with the Microsoft Azure AI stack., Development of POCs for different use cases using Generative AI and libraries like Langchain., Collaborated with a cross-functional team (AI/ML, DevOps, front-end/back-end) and USA leadership to support demos, and competitive analysis against market competitive tools.

Developed and optimized demand forecasting solutions for 71 retail outlets using historical sales data, seasonal trends, and market insights.
Generated accurate short-term and long-term forecasts to ensure optimal product availability and minimize excess inventory.
Collaborated with sales and production teams to refine forecasts based on promotions, market shifts, and new product launches.
Led stock-out analysis across all outlets, identifying root causes such as supply delays, sales spikes, and inventory gaps.
Designed and implemented a dynamic Excel dashboard to monitor stock levels, sales, and out-of-stock events in real time.
Automated manual reporting tasks using VBA macros, significantly improving operational efficiency and accuracy.

Developed and optimized demand forecasting solutions for 71 retail outlets using historical sales data, seasonal trends, and market insights., Generated accurate short-term and long-term forecasts to ensure optimal product availability and minimize excess inventory., Collaborated with sales and production teams to refine forecasts based on promotions, market shifts, and new product launches., Led stock-out analysis across all outlets, identifying root causes such as supply delays, sales spikes, and inventory gaps., Designed and implemented a dynamic Excel dashboard to monitor stock levels, sales, and out-of-stock events in real time., Automated manual reporting tasks using VBA macros, significantly improving operational efficiency and accuracy.

Designed and implemented supervised learning models across multiple use cases Movie Genre Classification based on metadata and textual descriptions, Spam SMS Detection using text preprocessing and classification and Credit Card Fraud Detection using anomaly detection and binary classification.
Conducted feature engineering on plot summaries, transaction patterns, and message content to enhance model accuracy.
Preprocessed and vectorized textual data using TF-IDF and CountVectorizer for optimal input to classification algorithms.
Evaluated model performance using precision, recall, F1-score, confusion matrices, and ROC-AUC to guide iterative improvements.
Tuned hyperparameters and applied cross-validation to improve generalization and reduce overfitting.
Compared model outputs to identify the most effective algorithm for each use case across diverse datasets.

Designed and implemented supervised learning models across multiple use cases Movie Genre Classification based on metadata and textual descriptions, Spam SMS Detection using text preprocessing and classification and Credit Card Fraud Detection using anomaly detection and binary classification., Conducted feature engineering on plot summaries, transaction patterns, and message content to enhance model accuracy., Preprocessed and vectorized textual data using TF-IDF and CountVectorizer for optimal input to classification algorithms., Evaluated model performance using precision, recall, F1-score, confusion matrices, and ROC-AUC to guide iterative improvements., Tuned hyperparameters and applied cross-validation to improve generalization and reduce overfitting., Compared model outputs to identify the most effective algorithm for each use case across diverse datasets.

Developed a deep learning pipeline to detect driver drowsiness using facial features and eye movement patterns.
Implemented Convolutional Neural Networks (CNNs) for real-time image-based feature extraction from video frames.
Used Support Vector Machines (SVMs) for comparative feature extraction and classification to benchmark traditional ML approaches.
Integrated EfficientNetB0 for lightweight and high-accuracy sleepiness detection on edge devices.
Conducted comparative analysis of CNN, SVM, and EfficientNetB0 models to evaluate performance across accuracy, latency, and resource efficiency.
Preprocessed image datasets including frame extraction, normalization, and augmentation to improve model robustness.
Validated models using confusion matrices, ROC curves, and precision-recall metrics to ensure reliable detection.

Developed a deep learning pipeline to detect driver drowsiness using facial features and eye movement patterns., Implemented Convolutional Neural Networks (CNNs) for real-time image-based feature extraction from video frames., Used Support Vector Machines (SVMs) for comparative feature extraction and classification to benchmark traditional ML approaches., Integrated EfficientNetB0 for lightweight and high-accuracy sleepiness detection on edge devices., Conducted comparative analysis of CNN, SVM, and EfficientNetB0 models to evaluate performance across accuracy, latency, and resource efficiency., Preprocessed image datasets including frame extraction, normalization, and augmentation to improve model robustness., Validated models using confusion matrices, ROC curves, and precision-recall metrics to ensure reliable detection.