Hi! I am Zobair
ML Engineer with over 3 years of hands-on experience in data analysis, machine learning implementation, and software development. I have worked mainly on computer Vision and LLM projects. My main focus was training models for production and deploying them on local servers. (edge device/ multi GPUs) My journey began with a background in automotive engineering, evolving into a passionate pursuit of artificial intelligence.
Projects I worked on
Architectural plans
I utilized machine learning for segmentation of individual objects within the plans. This involved not only segmentation but also data cleansing and engineering to enhance the accuracy and quality of the analysis. Additionally, I implemented a regeneration process capable of generating multiple variants based on specified conditions, adding a layer of flexibility to the project.
waste-material detection
In the waste management project, I implemented a system that segmented and classified various trash materials, sorting them into 18 distinct categories. This initiative aimed to estimate contamination levels and enable informed decisions before the waste was transported to the recycling center. The segmentation and classification process enhanced the efficiency of waste analysis and contributed to more effective recycling practices.
Advanced Rag systems
I was the main engineer on RAG pipelines for two different projects. According to customers' need, I worked both with local and cloud models.
The pipeline included top to bottom rag engineering from pre-processing documents of different formats to advanced retrieval to custom evaluation methods.
Sonography classification
In the realm of sonography classification, my role extended beyond model development. I provided mentorship to labelers and collaborated with professional doctors to ensure accurate data labeling. Leveraging this meticulously labeled dataset, I then trained a state-of-the-art classification model, contributing to the advancement of accurate and reliable sonography analysis.
Enhancing Waste Management Efficiency:
In the context of trash unloading, I incorporated depth estimation, optical flow, and disparity techniques to detect unloadings effectively. This approach utilized advanced computer vision methods to assess the spatial characteristics and movement patterns during the unloading process. The integration of depth estimation and optical flow contributed to accurate detection and monitoring, enhancing the overall efficiency and decision-making in waste management operations.
"Scaling Intelligence
Multi-GPU Training of State-of-the-Art Models on Cloud Platforms and In-Home GPU Rigs, with Proficiency in Training on Google Cloud, Microsoft Azure, and Storage Optimization
Optimizing Intelligence:
Quantizing and Achieving 45 FPS on Edge Devices for Efficient Object Detection"
Stable Diffusion
I have extensive experience with image generating models in both waste management and architectural plans projects. Employing state-of-the-art diffusion models, I explored both conditional and unconditional methods. This technology proved invaluable in creating realistic and diverse visual representations, enhancing the analytical capabilities of waste management assessments and architectural plan visualizations.
Large language models
I adeptly fine-tuned large language models on custom datasets, enhancing their performance for domain-specific tasks, such as natural language understanding and generation like CHATGPT.
Optimizing Operations
Implementing outlier detection on conveyor belts and motion monitoring for factory machinery ensures heightened efficiency and proactive maintenance. These techniques contribute to a seamless and reliable industrial workflow.
In addition to technical proficiency, I bring expertise in team management, conducting training for new hires, and mentoring both junior programmers and professional doctors in the intricate process of data labeling. This comprehensive skill set helps with the integration of technical solutions and effective collaboration within diverse project teams.
UNITBV
I undertook two years of automotive engineering within a four-year bachelor's degree at Transilvania University of Brașov (unitbv), transferring earned credits seamlessly to specialize in artificial intelligence at the IU (International University of Applied Sciences) Currently on a gap year, I'm gaining practical insights before concluding my undergraduate studies.
Coursera/Meta
Currently pursuing the Data Analyst Professional Certificate on Coursera from META, I'm advancing through foundational courses like "Foundations: Data, Data, Everywhere," "Prepare Data for Exploration," and "Analyze Data to Answer Questions." These courses collectively equip me with comprehensive data analysis skills, culminating in the hands-on "Google Data Analytics Capstone: Complete a Case Study."
Within the Machine Learning Engineer Professional Certificate, I completed a comprehensive set of certifications, including "Feature Engineering," "Launching into Machine Learning," "Machine Learning on Google Cloud," "Machine Learning Operations (MLOPS)," and "Production Machine Learning System." These credentials collectively demonstrate my proficiency in various facets of machine learning, emphasizing both theoretical understanding and practical implementation.
Internship
During my internship , I gained invaluable insights into state-of-the-art practices, fostering a deep understanding of computer vision technologies. This experience also served as my introduction to Python, a language I've been programming in for 2.5 years. Continuously applying Python in bug fixing and deploying solutions into production has honed my expertise in the language and its practical applications.