Mallemoina Gurudarpan

Machine Learning & Computer Vision Engineer

Building intelligent vision-based systems and exploring hybrid deep learning models at the intersection of research and real-world applications.

About Me

Passionate about AI, Computer Vision, and pushing the boundaries of innovation.

👋 Hi! I'm a Machine Learning and Computer Vision Engineer at Masterworks, where I develop intelligent vision-based systems and explore cutting-edge deep learning architectures.

🚀 My expertise spans from real-time object detection to generative modeling, bridging research and production-ready solutions. I specialize in Edge AI, MLOps and Generative Modeling

🎓 Education: B.Tech in Electronics and Communication Engineering from IIIT RGUKT RK Valley (2024)

📍 Location: Hyderabad, India

Programming Languages

Python, C, C++, Rust

ML/AI Frameworks

PyTorch, TensorFlow, YOLO, DeepStream SDK, NVIDIA TAO Toolkit

MLOps & Tools

Docker, Kubernetes, MLflow, Weights & Biases, Dask, CI&CD

Specializations

Machine Learning, Deep Learning, Geometric Deep Learning, Computer Vision, 3D Computer Vision, Generative Modeling, MLOps

Featured Projects

Innovative solutions at the intersection of AI and Computer Vision, showcasing real-world impact.

Real-Time Object Detection with DeepStream

Integrated multiple RTSP streams for real-time object detection and analytics. Optimized NVIDIA DeepStream pipelines with distributed training via PyTorch DDP across multiple GPUs, reducing training time by 35%. Utilized MLflow and Weights & Biases for experiment tracking, hyperparameter tuning, and performance visualization, ensuring scalable and reproducible MLOps workflows.

DeepStream YOLO RTSP Edge AI MLflow Weights & Biases Distributed Training MLOps

License Plate Recognition System

Built end-to-end detection and OCR pipeline using NVIDIA TAO Toolkit. Automated real-time number plate extraction from video streams.

TAO Toolkit OCR Computer Vision

Vision-Language Model Auto Annotation

Leveraged Grounding DINO for automatic annotation generation on unlabelled datasets, significantly reducing manual labeling effort.

VLM Grounding DINO Automation

Synthetic Data Generation Pipeline

Created synthetic datasets for license plate images and tabular time-series data using GANs. Enhanced model robustness through domain randomization, with distributed training via PyTorch DDP for faster generation. Leveraged MLflow and Weights & Biases for tracking experiments and optimizing hyperparameters, improving dataset quality by 30%.

Synthetic Data GANs Augmentation MLflow Weights & Biases Distributed Training MLOps

Hybrid Generative Modeling (GAN + EBM)

Developed a hybrid model combining GANs and EBMs for controlled image generation. Implemented distributed training with PyTorch DDP to scale across GPUs, reducing training time by 40%. Used MLflow and Weights & Biases for experiment tracking and visualization, ensuring robust and reproducible training pipelines.

GANs EBM Research MLflow Weights & Biases Distributed Training MLOps

Edge-based Action Detection

Deployed lightweight VLMs on edge devices for real-time action detection using RTSP video streams.

Edge AI VLM Real-time

Predictive Modeling for Operations

Developed statistical and Ensemble Models to predict off-block time probability (POBT) and vehicle inspection end times.

Time Series Stastical and Ensemble Models MLOps

Eye Spect Lens Defect Detection

Built a defect detection pipeline using YOLO-based vision models for high-precision quality inspection.

Quality Control YOLO Industrial AI

Mixture-of-Recursions with GAN for Synthetic Image Generation

Developed a novel generative model combining Mixture-of-Recursions with GANs to create high-quality synthetic images with enhanced diversity and realism. Utilized MLflow and Weights & Biases for comprehensive experiment tracking, hyperparameter optimization, and visualization of training metrics. Implemented distributed training via PyTorch DDP for efficient scaling across GPUs on both detection models (e.g., YOLO-based pipelines) and hybrid generative models, reducing training time by 40% while ensuring reproducibility in MLOps workflows.

GANs Mixture of Recussions Synthetic Images Generative AI MLflow Weights & Biases Distributed Training MLOps

Let's Connect

Open to discussing ML research, computer vision projects, or collaboration opportunities.

Feel free to reach out for collaborations, research discussions, or just to say hi! I'm always excited to connect with fellow AI enthusiasts.