Projects
🛡️ Malicious SQL Query Detection
Developed a security-focused system to detect malicious SQL queries before execution. The pipeline starts with lexical analysis to tokenize SQL inputs and extract meaningful patterns. Features such as keyword frequency, nesting levels, and suspicious patterns were engineered and fed into ML classifiers. Used labeled datasets of benign and malicious queries to train models. Achieved significant improvement in detection accuracy by combining lexical and statistical approaches.
Tech Stack: Python, Scikit-learn, Lexical Analysis, NLP, SQL, Data Preprocessing, Regex
🛡️ Malicious SQL Query Detection
Developed a security-focused system to detect malicious SQL queries before execution. The pipeline starts with lexical analysis to tokenize SQL inputs and extract meaningful patterns. Features such as keyword frequency, nesting levels, and suspicious patterns were engineered and fed into ML classifiers. Used labeled datasets of benign and malicious queries to train models. Achieved significant improvement in detection accuracy by combining lexical and statistical approaches.
Tech Stack: Python, Scikit-learn, Lexical Analysis, NLP, SQL, Data Preprocessing, Regex
🛡️ Malicious SQL Query Detection
Developed a security-focused system to detect malicious SQL queries before execution. The pipeline starts with lexical analysis to tokenize SQL inputs and extract meaningful patterns. Features such as keyword frequency, nesting levels, and suspicious patterns were engineered and fed into ML classifiers. Used labeled datasets of benign and malicious queries to train models. Achieved significant improvement in detection accuracy by combining lexical and statistical approaches.
Tech Stack: Python, Scikit-learn, Lexical Analysis, NLP, SQL, Data Preprocessing, Regex
☁️ Cloud-Based Solution for Residential Security and Efficiency
Built a serverless cloud system using AWS to automate visitor and vehicle entry tracking in gated communities. Used AWS Rekognition for OCR on vehicle plates, verified against a secure owner database in DynamoDB. Lambda functions handled real-time validation and event-triggered notifications via SNS. Added a mobile-accessible front-end interface to manage and approve visitors. Entire pipeline is scalable, cost-efficient, and integrated with security logs for audits. Improved gate efficiency and resident safety.
Tech Stack: AWS (Rekognition, Lambda, DynamoDB, SNS), Python, Cloud Architecture, Serverless Computing,
☁️ Cloud-Based Solution for Residential Security and Efficiency
Built a serverless cloud system using AWS to automate visitor and vehicle entry tracking in gated communities. Used AWS Rekognition for OCR on vehicle plates, verified against a secure owner database in DynamoDB. Lambda functions handled real-time validation and event-triggered notifications via SNS. Added a mobile-accessible front-end interface to manage and approve visitors. Entire pipeline is scalable, cost-efficient, and integrated with security logs for audits. Improved gate efficiency and resident safety.
Tech Stack: AWS (Rekognition, Lambda, DynamoDB, SNS), Python, Cloud Architecture, Serverless Computing,
☁️ Cloud-Based Solution for Residential Security and Efficiency
Built a serverless cloud system using AWS to automate visitor and vehicle entry tracking in gated communities. Used AWS Rekognition for OCR on vehicle plates, verified against a secure owner database in DynamoDB. Lambda functions handled real-time validation and event-triggered notifications via SNS. Added a mobile-accessible front-end interface to manage and approve visitors. Entire pipeline is scalable, cost-efficient, and integrated with security logs for audits. Improved gate efficiency and resident safety.
Tech Stack: AWS (Rekognition, Lambda, DynamoDB, SNS), Python, Cloud Architecture, Serverless Computing,
🎨 Realistic Product Visualization Using DreamBooth & Hugging Face
During my internship, I developed a generative AI pipeline to convert raw product samples into realistic, human-worn visuals for marketing and branding purposes. The workflow involved fine-tuning a Stable Diffusion model using DreamBooth with company-provided sample images. Guided prompts were designed to control visual attributes like pose, lighting, and context. This solution enabled the generation of high-fidelity synthetic visuals, eliminating the need for physical photoshoots and accelerating marketing asset production.
Tech Stack: Stable Diffusion, DreamBooth, Hugging Face Diffusers, Prompt Engineering, Python, Gradio UI, Image Augmentation
🎨 Realistic Product Visualization Using DreamBooth & Hugging Face
During my internship, I developed a generative AI pipeline to convert raw product samples into realistic, human-worn visuals for marketing and branding purposes. The workflow involved fine-tuning a Stable Diffusion model using DreamBooth with company-provided sample images. Guided prompts were designed to control visual attributes like pose, lighting, and context. This solution enabled the generation of high-fidelity synthetic visuals, eliminating the need for physical photoshoots and accelerating marketing asset production.
Tech Stack: Stable Diffusion, DreamBooth, Hugging Face Diffusers, Prompt Engineering, Python, Gradio UI, Image Augmentation
🎨 Realistic Product Visualization Using DreamBooth & Hugging Face
During my internship, I developed a generative AI pipeline to convert raw product samples into realistic, human-worn visuals for marketing and branding purposes. The workflow involved fine-tuning a Stable Diffusion model using DreamBooth with company-provided sample images. Guided prompts were designed to control visual attributes like pose, lighting, and context. This solution enabled the generation of high-fidelity synthetic visuals, eliminating the need for physical photoshoots and accelerating marketing asset production.
Tech Stack: Stable Diffusion, DreamBooth, Hugging Face Diffusers, Prompt Engineering, Python, Gradio UI, Image Augmentation
🧠 Multimodal Brain Tumor Segmentation with Explainable AI
Developed and compared deep learning models (ResAttUNet, DeepLabV3+, U-Net with BesNet) to perform accurate brain tumor segmentation using multimodal MRI slices (FLAIR, T1, T1ce, T2). Integrated Grad-CAM and Integrated Gradients to visualize model focus, improving interpretability and clinical relevance. Achieved a highest Dice Score of 0.8448 with DeepLabV3+ and spatial precision using ResAttUNet.
Tech Stack: Python, PyTorch, ResAttUNet, DeepLabV3+, U-Net, BesNet, Grad-CAM, Integrated Gradients, NumPy, OpenCV, Matplotlib, Medical Image Preprocessing (NIfTI), Dice Score Evaluation
🧠 Multimodal Brain Tumor Segmentation with Explainable AI
Developed and compared deep learning models (ResAttUNet, DeepLabV3+, U-Net with BesNet) to perform accurate brain tumor segmentation using multimodal MRI slices (FLAIR, T1, T1ce, T2). Integrated Grad-CAM and Integrated Gradients to visualize model focus, improving interpretability and clinical relevance. Achieved a highest Dice Score of 0.8448 with DeepLabV3+ and spatial precision using ResAttUNet.
Tech Stack: Python, PyTorch, ResAttUNet, DeepLabV3+, U-Net, BesNet, Grad-CAM, Integrated Gradients, NumPy, OpenCV, Matplotlib, Medical Image Preprocessing (NIfTI), Dice Score Evaluation
🧠 Multimodal Brain Tumor Segmentation with Explainable AI
Developed and compared deep learning models (ResAttUNet, DeepLabV3+, U-Net with BesNet) to perform accurate brain tumor segmentation using multimodal MRI slices (FLAIR, T1, T1ce, T2). Integrated Grad-CAM and Integrated Gradients to visualize model focus, improving interpretability and clinical relevance. Achieved a highest Dice Score of 0.8448 with DeepLabV3+ and spatial precision using ResAttUNet.
Tech Stack: Python, PyTorch, ResAttUNet, DeepLabV3+, U-Net, BesNet, Grad-CAM, Integrated Gradients, NumPy, OpenCV, Matplotlib, Medical Image Preprocessing (NIfTI), Dice Score Evaluation
"He consistently exceeds our expectations"
"I recommend Goran whole-heartedly"
"Loved to work with Goran!"
"He consistently exceeds our expectations"
"I recommend Goran whole-heartedly"
"Loved to work with Goran!"
"He consistently exceeds our expectations"
"I recommend Goran whole-heartedly"
"Loved to work with Goran!"