Freelance Python & Machine Learning Engineer - Self-Employed - Remote, Islamabad, Pakistan
(2020-01)
- Designed and trained machine learning models to predict customer churn on structured datasets, achieving 84% AUC. This helped clients identify and retain at-risk customers.
- Built LSTM and Prophet-based time-series forecasting models for retail sales, aiding accurate inventory planning and demand forecasting for various clients.
- Used LSTM and BERT architectures for product-review sentiment classification, automating large-scale customer feedback analysis and providing actionable product insights.
- Extracted and compiled targeted customer lists from property websites using BeautifulSoup and Selenium, generating qualified leads for a real estate firm.
- Developed chatbots with TensorFlow/Keras, created interactive Streamlit and Plotly dashboards, and built classification/forecasting models (XGBoost, Random Forest, neural nets) for different clients.
Contract: CNN-Based Pest Identification System - Field-Applicable Pest Control Platform
(2022 - 2023)
- Developed and deployed a production CNN system that improved real-time pest detection accuracy by 22% over baseline models.
- Curated and processed over 12,000 field images, addressing class imbalance and quality issues to create a robust training dataset.
- Designed and fine-tuned four architectures (ResNet50, MobileNetV2, VGG16, and a custom lightweight CNN optimized for edge deployment).
- Reduced false positives by 35% through careful data-leakage mitigation and class-specific fine-tuning, achieving 94.7% accuracy on production test sets.
Contract: End-to-End Real Estate Data Pipeline & Analytics - Client Market-Analytics System
(2023 - 2024)
- Developed a GitHub Actions workflow that scraped over 15,000 real estate listings hourly while following rate limits and compliance rules.
- Built a Python (Pandas/NumPy) cleaning pipeline that removed duplicates, missing values, and inconsistent formats, cutting data errors by 40%.
- Deployed a real-time Flask analytics dashboard that provides pricing trends, geographic heatmaps, and predictive market insights with less than 2 seconds of latency.
- Maintained 98% pipeline uptime and reduced client teams' manual data-entry workload by 80% after full production integration.
Contract: Product Review Sentiment Analysis Platform - Business NLP Solution
(2024 - 2025)
- Built end-to-end NLP classification pipelines to predict positive or negative product reviews from business datasets.
- Cleaned and analysed review text, generating word-cloud visualizations to highlight the most common positive and negative terms.
- Trained and compared models: Dense Keras (89% train / 50% test) and LSTM Keras (99% train / 83% test accuracy).
- Delivered a Streamlit application that provides top product analytics and real-time sentiment prediction for stakeholders.
