rockenzo

Description

👋 Hello there, I'm Rock! 🙋

Welcome to my profile! As an aspiring Software Engineer with a solid foundation in web development, software engineering, and machine learning, I have a passion for solving complex problems and building scalable applications. Eager to collaborate with diverse teams, I aim to contribute to impactful projects and push the boundaries of tech.

🛠 Skills 👨‍💻

1. 🔧 Software Development 📱

Languages:
Frameworks:

2. 🌐 Web Development 💻

Frontend:
Backend:
Databases:
Tools:
3. 🧠 Machine Learning 🤖
Frameworks:
Experience: Neural network training, optimization, predictive modeling, computer vision.

4. 📊 Data Analysis 🧐

Libraries:
Experience: Data manipulation and analysis.

5. 💬 LLM APIs and Model Localization 😶‍🌫️

APIs:
Transformers:
Platforms:
Experience: Proficient in utilizing LLM APIs for integrating conversational AI into applications, fine-tuning transformer models (BERT/DistilBERT) for domain-specific tasks, and localizing models using platforms like Ollama for on-device or private cloud deployment.

6. 📋 Project Management 🗂️

Methodologies: Agile, Scrum
Experience: Leading projects, coordinating teams, meeting deadlines.

🛠 Project Experience 📑

Engineered an advanced web scraper leveraging LLM Ollama gpt-oss:20b for intelligent data parsing, extracting and processing content from over 500 web pages on-premise with enhanced precision and performance.
Developed a responsive Streamlit interface with real-time progress tracking, sidebar configuration, and visual indicators for seamless user experience in URL input, scraping, and data parsing.
Implemented smart content cleaning and intelligent chunking algorithms using BeautifulSoup and lxml, significantly reducing data extraction errors and improving content processing efficiency.
Architected modular code structure with enhanced error handling, logging capabilities, export functionality, and configurable processing parameters for robust and reliable operation.

Architected and developed a comprehensive spatial data pipeline for an accessibility-focused mapping application serving Sydney, processing 76,000+ data points across 15+ datasets including building complexes, transportation networks, and accessibility infrastructure.
Implemented multi-service Docker architecture using Docker Compose to orchestrate Elasticsearch, PostgreSQL (GNAF), Redis, and Flask services, enabling scalable data processing and real-time spatial data visualization.
Engineered automated data indexing system with Elasticsearch, supporting both point and polygon geometries, with dynamic mapping configurations for heterogeneous spatial datasets (GeoJSON, EPSG4326 projections).
Developed real-time API integration pipeline using APScheduler for hourly pedestrian count data updates from ArcGIS REST services, implementing robust error handling and data transformation workflows.
Built Flask-based microservices architecture including GNAF address search API with Celery/Redis task queuing, Elasticsearch search API with geospatial filtering, and live bus tracking indexer for dynamic transportation data.
Optimized spatial query performance through bounding box filtering, bulk indexing strategies, and efficient Elasticsearch mappings, reducing query response times and enabling responsive interactive map interfaces.

Developed an advanced multiclass fraud detection system using BERT, DistilBERT, and FLAN-T5 transformers, achieving 97% accuracy in classifying 9 distinct fraud types including phishing, tech support scams, job scams, reward scams, and SSN scams alongside legitimate messages.
Engineered unified classification + reasoning pipeline using fine-tuned FLAN-T5 models for explainable fraud detection, generating both predictions and detailed explanations in a single inference pass with zero API costs through Kaggle GPU infrastructure.
Implemented production-ready inference demos with interactive Jupyter notebooks, command-line utilities, and batch processing capabilities, featuring confidence scoring, multiclass prediction, and comprehensive evaluation metrics for cybersecurity applications.
Optimized DistilBERT implementation for 60% faster training while maintaining 97% of BERT’s performance, with 40% smaller model size and memory-efficient processing ideal for real-time fraud detection and resource-constrained deployments.
Architected transformer-based preprocessing pipeline with stratified sampling, tokenization optimization, and GPU-accelerated training on Kaggle, enabling scalable deployment with automated model checkpointing and early stopping for enterprise fraud detection systems.
Built comprehensive model comparison framework analyzing multiple LLM architectures (GPT-2, BERT variants, T5 models) with performance visualization, statistical analysis, and detailed reasoning quality assessments for cybersecurity research applications.

Conceptualized and built a web application to boost visibility and streamline administrative functions for volunteer groups, featuring user sign-ups, location management, event creation, and role-based profile management.
Spearheaded the database schema design and led the development of the user management module, ensuring efficient data storage, retrieval, and seamless frontend integration for effective role and access management.
Implemented secure authentication mechanisms using Argon2 for password hashing and Google OAuth for third-party login.
Coordinated with frontend and backend teams to achieve a 20% reduction in development time ahead of schedule.
Performed comprehensive testing and debugging to ensure high performance and user satisfaction.

Created and honed a model for detecting football players in videos with the YOLOv5 framework for training and YOLOv8 for inference, achieving 92% accuracy.
Sourced and pre-processed a dataset of 663 annotated images from Roboflow, enhancing model precision.
Leveraged Google Colab’s GPU resources for efficient model training and tuning, reaching 97.5% precision in generating real-time bounding boxes around detected players.
Minimized manual video analysis time by 85% through model performance optimizations.

Developed a weather application using SwiftUI, providing real-time weather updates based on user location.
Integrated CoreLocation to fetch and manage user location data for accurate weather information.
Implemented a responsive UI with animated loading and launch screens to enhance user experience.
Utilized the OpenWeather API to fetch reliable and up-to-date weather data.
Designed custom views and extensions for streamlined UI components and improved code reusability.
Conducted thorough testing and debugging to ensure application stability and performance.

Developed a semantic segmentation pipeline for autonomous driving scenes using PyTorch.
Implemented custom data loaders, augmentation strategies, and advanced network architectures to boost segmentation accuracy.
Conducted ablation studies and tracked results with mIoU and FLOPs metrics for model efficiency.
Documented findings and methods in a comprehensive report, highlighting improvements over the baseline.

Engineered a reinforcement learning agent using Deep Q-Networks (DQN) to autonomously play Flappy Bird, achieving high scores and advancing through increasing difficulty levels.
Refined the agent’s state representation and reward function, leveraging game environment parameters (e.g., bird position, pipe gap, velocity) to optimize learning and decision-making.
Implemented experience replay and target network updates for stable and efficient training, following best practices from modern RL research.
Utilized PyTorch for neural network modeling and training, ensuring scalable and reproducible experiments.
Validated agent performance across multiple game levels, with automated evaluation and model checkpointing for robust benchmarking.

🎓 Education 📖

Bachelor of Information Technology: February 2023 – Present
University of Adelaide

Relevant coursework:
- Artificial Intelligence Technologies (Distinction)
- Computer Systems
- Object-Oriented Programming
- Programming for IT Specialists (Distinction)
- Web & Database Computing (Distinction)
- Introduction to Applied Programming (High Distinction)
- Introduction to Computer Systems, Networks, and Security (High Distinction)
- Database and Ethical Data
- Secure Programming
- Using Machine Learning Tools
- Computer Vision

📜 Certification 🏆

HackerRank Software Engineer Certificate
Credential ID: 1D18ED03135B
- Skills: Problem-solving, Python programming, SQL databases, REST APIs
- Issued: September 2024