About Me
Data Scientist and Machine Learning Engineer with 2 years of experience building computer vision, NLP, and forecasting systems for production environments. Delivered a 35–50% reduction in manual inspection effort through…
Data Scientist and Machine Learning Engineer with 2 years of experience building computer vision, NLP, and forecasting systems for production environments. Delivered a 35–50% reduction in manual inspection effort through automated defect detection, and cut model training resource requirements by 75% through custom fine-tuning. Strong foundation in Python, SQL, deep learning, and cloud deployment (AWS, Docker). Currently based in Fujairah and looking to bring this experience to a data-driven engineering team in the UAE.
Experience
Data Science Associate
<p>Designed an automated optical inspection system to verify PCB component presence, position, and orientation using factory inspection cameras.</p><p> Trained a custom YOLOv8 image model for component segmentation and classification, detecting missing, misplaced, and incorrect components on PCB images.</p><p> Applied polarity checks and coordinate-based validation to ensure correct component orientation and alignment with reference layouts.</p><p> Automated accept/reject decisions for PCB boards and logged inspection results by serial number for full traceability and quality assurance.</p><p> Built a testing environment for the team using Streamlit and Docker.</p><p> Developed a custom TTS model and audio chatbot using the Qwen3 TTS Base Model for natural voice generation.</p><p> Engineered a memory-efficient fine-tuning approach, reducing VRAM requirements from 48GB to 12GB (75% reduction) without compromising output quality.</p><p> Built a data preprocessing pipeline covering audio collection, noise filtering, and resampling to 24kHz to enhance speech clarity and model performance.</p><p> Performed model fine-tuning and inference testing to ensure realistic, consistent voice synthesis.</p><p> Containerized the solution with Docker for reproducible team testing and deployed it on AWS for scalable, real-time usage.</p><p> Built an automated defect detection system for semiconductor chip cuts using OpenCV (image segmentation and contour detection).</p><p> Integrated micro ArUco markers for high-precision dimension measurement and calibration.</p><p> Deployed with industrial Teledyne DALSA and DKP cameras (130x zoom), achieving real-time processing at 24 FPS on low-end systems.</p><p> Achieved ±0.1mm measurement accuracy, enabling reliable detection of defective chip cuts and reducing manual inspection effort by 35–50%.</p><p> Designed a batch-level defect logging system to improve traceability and quality monitoring across manufacturing workflows.</p><p> Built a Python-based application to accurately extract and structure invoice details from PDFs using NLP and regex techniques.</p><p> Developed a front-end interface for invoice data extraction with automated push to a MySQL database.</p><p> Extracted data using the pdfplumber library combined with custom regex patterns.</p><p> Designed predictive models using LSTM and XGBoost to forecast blood glucose levels from time-series health data.</p><p> Implemented RNN, LSTM, and GRU-LSTM models for time-series forecasting of the Nifty 50 index, analyzing key trend drivers across stock categories and building a 7-day price prediction model to support buy/sell decision-making.</p><p> Designed and implemented a pipeline to fine-tune the DeepSeek base model using Supervised Fine-Tuning (SFT) combined with human feedback.</p><p> Utilized Unsloth to accelerate inference and training, and researched fine-tuning strategies to develop a combined SFT and human-feedback framework.</p>
Data Science Intern
<p>Developed an NLP-based recommender system for user profile search in collaboration with the AI team.</p><p> Evaluated and recommended data analysis tools to address key business problems.</p><p> Cleaned and manipulated raw datasets to support downstream analysis and modeling.</p>