Technical Skills

  • Cybersecurity Software: Metasploit, Burp Suite, OWASP-ZAP, Amass, Nmap, Wireshark, SQLmap, John the Ripper, Hydra, Sublist3r, dirbuster, goBuster, Nessus, Snort, recon-ng, Hydra, OpenVAS, BeEF, Nikto, Webscarab, Shodan
  • Programming: Python, Perl, PHP, Shell, and TCL (good), MySQL, and SQL (intermediate), JavaScript, C, C++, Ruby, and PowerShell (basic)
  • Operating Systems & Tools: Kali Linux, Windows, Linux, MacOS, AWS (Intermediate), GCP, and Git (Basic)
  • GenAI LLM: Hugging Face, LangChain, Multimodal RAG, OpenAI, BERT, RoBERTa, Flan-T5, Salesforce BLIP, GCP Models
  • ML Models: Keras, Sci-Kit learn, Pandas, Numpy, PyTorch, Tensorflow

Ethical Hacking Related Notes/Projects & Bug Bounty Competitions

Penetration Testing against vulnerable apps on Oracle VirtualBox

• Reconnaissance and Fingerprinting using OSINT tools; Shodan, Google Dorks, Burp & ZAP passive scanning.
• Network Ports mapping and scanning using Nmap, Masscan. Network data packets sniffing using Wireshark.
• Passive and Active scanning; manual & automated, spidering, directory brute-force, etc. (Burp Suite, OWASP-ZAP, Nessus)
• Injection techniques to exploit Web apps: XSS, CSRF, SSRF, SQL, brute-force passwords and users, HTML Headers, etc. (Burp)
• Analyzed system for potential vulnerabilities from improper system and network configuration – Automated and Manual
• Exploitation, and Post exploitation using Metasploit
GitHub

GenAI LLM & Deep Neural Networks - Work Projects & Competitions

Trained from scratch Huggingface ROBERTaMLM Model

Created a config to decrease number of parameters to 84m and trained on Google Colab against oscar.eo.txt file (approx. 1 million lines), picked 10k lines from file. Split the file into train, eval, and test sets. For tokenization used ByteLevelBPETokenizer, BertProcessing, and RobertaTokenizerFast. Due to machine resources trained on 10 epochs.
GitHub

Pretraining Google Flan-T5-small against ‘samsum’ dataset

Pretraining Google Flan-T5-small against ‘samsum’ dataset using PEFT LoRA, an Prompt instructions and verify ROGUE and BLUE scores against test dataset
GitHub

Pretrain and generate image captions using Salesforce blip model

Pretrain and generate image captions using Salesforce blip model against ‘h-and-m-fashion-caption’ dataset and generate text related to image in test dataset and check accuracy scores
GitHub

Streamlit App zilliar_chat

Streamlit App zilliar_chat using Langchain and OpenAI model. App requires User access token to enable App prompt. Implemented App to include Agent Chain (this is first version – working on enhancements)
GitHub

Explainability of DLNN limitation on a Multiclass Classification Dataset

Evaluated Multiclass Classification dataset (Obesity Risk) to provide a comprehensive understanding of deep learning fully connected Neural Network behavior when a dataset contains many categorical columns and most of which are binary. Conducted a thorough validation of model performance using various scaling, encoding, Decision Trees, and PCA techniques, conclusion is that DNNs decisions are biased. Thorough investigation is needed to better understand such behavior projected by DNNs.
GitHub    Kaggle

34-layer ResNet - predict the effect of vasculature "flow of blood, oxygen"

Designed and implemented a 34-layer Residual Deep Neural Network similar to ResNet50 for Kaggle competition on 3D multiresolution imaging datasets of kidneys for SenNet + HOA – Hacking the Human Vasculature in 3D to predict the effect of vasculature "flow of blood, oxygen" in the human body through the vessel network.
GitHub    Kaggle

NLP - Capstone project - SwiftKey

Exploratory analysis done to achieve the goal towards an eventual app on Shiny, and certification as a Data Science Specialist awarded by John Hopkin’s University. The corpus data is provided by SwiftKey’s Natural Language Processing (NLP) project. Created bigrams, trigrams, and quad-grams to predict next word    Presentation
RPubs    GitHub    Shiny App

Home Prices Prediction - King County, Washington State

The dataset was provided during "Machine Learning Specialization" certification course offered by Univ of Washington, and taught by: -- Emily Fox, Amazon Professor of Machine Learning -- Carlos Guestrin, Amazon Professor of Machine Learning
RPubs    GitHub    Shiny App

Hyperparameters Fine-Tuning - A Deep Dive

Hands-on experience in fine tuning hyper parameters for Deep Neural Networks using Keras_Tuner, Bayesian Optimization, Random Search techniques.
GitHub    Kaggle

Predict Cardiovascular disease risk in individuals with high obesity

Designed and implemented Fully connected Neural Nets to Predict Cardiovascular disease risk in individuals with high obesity
GitHub    Kaggle

Binary Classification with a Bank Customer Churn Rate

Designed a fully connected Deep Neural Network to predict whether a customer continues with their account or closes them for Binary Classification with a Bank Churn Dataset.
GitHub    Kaggle

Exploratory Data Analysis

EDA analysis on Obesity dataset to improve accuracy for Deep Neural Networks. If the dataset is not normalized - no skew, then Neural Nets prediction is high, else the results are stuck in a valley
GitHub    Kaggle