Hi, I'm Mohamad Zamini.
A
Self-driven and passionate about advancing machine learning, I am focused on pushing the boundaries of Large Language Models (LLMs) through innovative research and development. As a final-year PhD Candidate with hands-on experience in optimizing LLMs during my recent internship, I am committed to solving complex real-world problems with cutting-edge AI technology. My ambition is to contribute to the future of AI by developing scalable and efficient models that can transform industries and enhance human-computer interaction.
About
I am a PhD candidate in Computer Science with a focus on optimizing Multimodal Large Language Models (MLLMs) to enhance their reasoning capabilities. My work involves accelerating LLMs through advanced techniques such as pruning, ensuring performance is maintained or improved. I have hands-on experience with foundational models, having previously interned at Numenta Inc., and I am currently developing innovative approaches like Mixture of Depth (MoD), Mixture of Experts (MoE) for resamplers, and attention pruning to push the boundaries of MLLM efficiency and scalability.
- Programming Languages: Python, C++
- Databases: MySQL, MongoDB, PostgreSQL
- Libraries & Frameworks: PyTorch, TensorFlow, Hugging Face Transformers, Keras, NumPy, Pandas, OpenCV
- Model Optimization & Deployment: ONNX, TensorRT, TorchServe, FastAPI
- Tools & Platforms: Git, Docker, Kubernetes, AWS, GCP, Azure, JIRA, Weights & Biases (wandb)
Seeking a challenging position that leverages my expertise in Machine Learning and Software Engineering, offering opportunities for professional development, innovative experiences, and personal growth.
Experience
- Built LLM-driven analytics agent enabling natural-language querying and multi-turn reasoning over large-scale telemetry logs.
- Automated weekly monitoring of retention and engagement metrics using SHAP, ANOVA, and anomaly-aware delta detection.
- Designed scalable data pipelines supporting session-level behavioral modeling.
- Tools: Python, PyTorch
- Fine-tuned Mistral and LLaMA models with activation sparsity and attention sparsity for efficient inference.
- Developed dynamic context pruning, KWTA mechanisms, and KV caching optimizations.
- Tools: Python, PyTorch, Accelerate, GPT, llama
- Designed semantic compression system using deep autoencoders for high-dimensional data.
- Built ML models for geothermal data analysis and improved accuracy through algorithmic optimization.
- Fine-tuned BART for Persian text summarization.
- Implemented Matrix Factorization for topic modeling.
- Implemented BiLSTM-CRF for sequence tagging and matrix factorization for topic modeling.
- Tools: Python, Scikit-learn, NLTK
Projects
SHAP vs Lime Vs ELI5
- Tools: Python, PyTorch
- To explain the model's predictions, the project uses model interpretability tools such as SHAP (SHapley Additive exPlanations), Lime (Local Interpretable Model-agnostic Explanations), and Eli5 (Explain Like I'm 5). These tools provide insights into how the model makes decisions and highlight the importance of different features in predicting strokes..
Skills
Languages and Databases
Python
HTML5
CSS3
MySQL
PostgreSQL
Shell Scripting
Libraries
NumPy
Pandas
OpenCV
scikit-learn
matplotlib
NLTK
Frameworks
Django
Flask
Bootstrap
Keras
TensorFlow
PyTorch
Other
Git
AWS
Docker
Education
Laramie, WY
Degree: PhD in Computer Science
Area of Study: Causal Reasoning for Improving Generalization in Visual Question Answering
- Intro to Artificial Intelligen
- Machine Learning
- High Perform Comput & Paradigm
- Advanced Image Processing
- Neural and Fuzzy Systems
Relevant Courseworks:
Tehran, Iran
Degree: Masters of Information Technology
CGPA: 3.68/4.0
- Artificial Neural Networks
- Neural and Fuzzy Systems
Relevant Courseworks: