Problem → approach → result. Selected work across deep learning, computer vision, NLP and production data engineering.
Problem — Classify 10 land-use categories from Sentinel-2 satellite imagery.
Approach — EfficientNetB0 transfer learning with targeted augmentation instead of training a CNN from scratch.
Result — Strong test accuracy across all 10 classes, validating transfer learning for a mid-sized remote-sensing dataset.
Problem — Beat a teacher-provided ResNet50 baseline (92.22% test accuracy) on tumor classification from MRI scans.
Approach — Built an EfficientNetB0 pipeline with preprocessing tuned for medical imaging noise and class imbalance.
Result — Outperformed the deeper ResNet50 baseline — architecture choice mattered more than depth here.
Problem — selenium-wire breaks on Python 3.12, blocking authenticated residential-proxy scraping at scale.
Approach — Built a Chrome-extension-based proxy auth layer, decoupling proxy auth from the driver.
Result — A product scraper and link-checker running reliably without the usual version lock-in.
Problem — Identify plant diseases from leaf images to support early intervention.
Approach — Trained a CNN from scratch in TensorFlow/Keras with augmentation and iterative evaluation.
Result — A working classifier covering the full pipeline — preprocessing, augmentation, training, evaluation.
Problem — Collect, clean and structure large-scale text data for language model experimentation.
Approach — Built a text-processing pipeline covering collection, preprocessing and analysis.
Result — A reusable NLP pipeline foundation for further language-generation experiments.
Problem — Detect and classify objects in images in real time.
Approach — Prepared and labeled a custom dataset, then trained and evaluated a YOLO-based model.
Result — A functioning real-time detection system, benchmarked on accuracy and inference speed.
Problem — Ingest, transform, and load large volumes of data from diverse sources.
Approach — Designed and implemented a scalable ETL pipeline using Python and Apache Airflow.
Result — A robust pipeline that reduced data processing time by 60% and improved data quality.
Problem — Create a modern, responsive website for a digital agency.
Approach — Designed the UI/UX and implemented the frontend using HTML, CSS, and JavaScript.
Result A visually appealing and functional website that effectively showcases the agency's services.
Website Link https://scrapixai.com
Problem — Scrape data from different types of Ecommerce websites using Python and Selenium.
Approach — Build a Web Scraping Bot having different files run.py, config.py,browser.py, scraper.py and requirements.txt
Result Run the bot to scrape data from different types of Ecommerce websites also integrated Proxy Rotation and Anti-Detection Measures
Website Link Bol.com , Triveturia.com , Amazon.ca.com , Instagram profiles and many more Ecommerce websites