Javad Kasravi

- Fine-tuned a deep neural network with 70 billion parameters using full precision for an economic game, employing model and data parallelization with only 8 GPUs - Employed state-of-the-art compression techniques (quantization, LoRA, and QLoRA,...) to scale ML models - Created isolated environments using Docker and Singularity, and applied it for multinode training - Applied the Slurm workload manager within an HPC cluster environment for the purpose of fine-tuning LLM

- Developed and fine-tuned 3D-cycleGANs, Contrastive Unpaired image to image Traslation, StyleGAN2-ADA, Swin Trasformer model to generate synthetics images - Evaluated and tested different algorithms for denoising and enhancing the quality of images - Skilled in coordinating multi-node training using distributed computing frameworks

• Processed time-series data to extract earthquake signals using deep learning • Prepared, cleaned, and processed a large amount of time-series data to feed into supervised and unsupervised learning algorithms • Evaluated Density-based clustering and Bayesian Gaussian mixture model algorithm to identify earthquake events • Analyzed and applied different deep learning algorithms (CNN and RNN) using PyTorch

- Developed and visualized drilling dashboards to optimize drilling operation using business intelligence tools, resulting in dropping operations cost by 10%. - Led drilling engineers and provided input to the competency development plans for the engineers - Extracted forthcoming challenges during drilling operations using a huge amount of data aligned with statistics - Proposed programs according to gathering and screening structured and instructed data

- Introduced text mining for daily drilling Reports to extract statistics and pattern - Wrote drilling failure report according to existing time series data and created a loop to revise the decisions

• Wrote and developed a MATLAB interface toolbox for coupling FLAC3D (finite difference method) and ECLIPSE Simulation (finite volume method) • Guided several MSc students to build Geomechanical Models

• Wrote and developed a MATLAB interface toolbox for coupling FLAC3D (finite difference method) and ECLIPSE Simulation (finite volume method) • Guided several MSc students to build Geomechanical Models

• Prepared and reviewed commodities and spare parts inventory database (SPID), approved their documents and announced to builders, and coordinated with project managers

• Thesis Topic: “Optimization of Vertical and Horizontal Wells Design in Gas Reservoirs Using Genetic Algorithm Approach.” (Under financial Support from National Iranian Central Oil Fields Company). • Wrote and developed a user-friendly MATLAB optimization toolbox • Coupled MATLAB optimization toolbox with ECLIPSE Simulation (Finite Volume Method) • Develop and fully automated artificial intelligence optimization engine to select the best scenarios • Total GPA: 17.57/20.00