Arashdeep Mehroke

Machine Learning Engineer | Data Scientist

Solving real-world problems with AI, cloud infrastructure, and big data. With 3+ years of experience and a Master's in progress at Georgia Tech, I build intelligent systems that scale.

View My WorkGet in Touch
Minimalistic Data Science Portrait
Clean Code
ML Systems
Cloud Native

About Me

I'm deeply passionate about data science and machine learning—fields that constantly challenge me to think critically, build creatively, and solve meaningful problems. I'm currently pursuing a Master's in Data Science at Georgia Tech to deepen my understanding, while continuously working on hands-on projects ranging from real-time emotion recognition to computer vision and synthetic data generation.

I thrive on learning by doing, whether it's building full-stack applications or optimizing machine learning pipelines. My technical toolkit includes Python, SQL, Java, TensorFlow, PyTorch, Scikit-learn, Docker, Kubernetes, AWS, GCP, Pandas, NumPy, Tableau, Power BI, and ArcGIS.

Python
SQL
C++
TensorFlow
PyTorch
Scikit-learn
XGBoost
AWS
GCP
Docker
Kubernetes
Git
CI/CD
Flask
Django
React
Node.js
Power BI
ArcGIS

Featured Projects

Emotion Recognition Using RAVDESS Audio Data
System for recognizing human emotions from audio using the RAVDESS dataset. Includes feature extraction (YAMNet, Librosa), dataset balancing (SMOTE), dimensionality reduction (t-SNE, UMAP), and model training for robust emotion classification.

Tech Stack: Python, TensorFlow, Librosa, SMOTE, UMAP, t-SNE

GitHub
Real-Time Theft Detection System
System for detecting retail theft using live video streams from IP cameras routed through Raspberry Pi 5 and AWS. Covers object detection (YOLOv8), cloud video analysis (Rekognition), edge inference (TensorFlow Lite), and real-time notifications (SNS). Designed for mobile-accessible, cost-effective surveillance.

Tech Stack: Raspberry Pi 5, Python, TensorFlow Lite, OpenCV, AWS Kinesis, Rekognition, IoT Core, SNS, API Gateway

GitHub
Synthetic Retail Store Dataset with Unity
Simulated retail store environment using Unity to generate synthetic training data for retail computer vision models. Features smart viewport-based labeling and IoU filtering to reduce noisy annotations, enhancing YOLOv8 fine-tuning and behavior detection (e.g., theft via PoseLift).

Tech Stack: Unity, Python, YOLOv8, OpenCV, NumPy, PoseLift, IoU Filtering

GitHub

Professional Experience

Machine Learning Engineer

Office of Performance and Data Analytics • May 2024 - Present

  • Developed and optimized PyTorch-based deep learning models to enhance urban safety initiatives
  • Designed and deployed ETL pipelines using Azure Databricks, improving efficiency and ensuring data integrity
  • Created an interactive dashboard for real-time analytics on city data, enhancing accessibility and decision-making
  • Deployed AI models using AWS SageMaker, S3, and Lambda for scalable, cloud-based inference
  • Built a large-scale feature engineering framework, increasing predictive model accuracy by 15%

Technical Lead

Western Digital • January 2023 - May 2024

  • Utilized Python and OpenCV to generate synthetic silicon wafer images with defects for model training
  • Improved defect classification models by increasing dataset diversity, boosting detection accuracy by 25%
  • Designed and implemented a data lake using S3 for efficient storage and retrieval of image data
  • Optimized distributed model training and hyperparameter tuning with Apache Spark

Web Application Developer

Sierra Nevada Research Institute • January 2022 - December 2022

  • Developed a dynamic front-end application using React and JavaScript
  • Implemented a Django-based backend for data storage, ensuring ACID compliance
  • Designed and deployed RESTful API endpoints for efficient data serving

Data Analyst

Blu-Lite Inc. • May 2018 - December 2021

  • Developed regression models to forecast sales and inventory trends, driving a 5% increase in revenue
  • Created dashboards and visual reports with Power BI and Excel, improving data accessibility by 25%
  • Automated routine data reporting processes, reducing manual effort by 20%

Get in Touch

Download Resume