Atharva Pore

Experience

• 

  IUCAA, Pune, India

  Dec 2019 - Jun 2023

  Software Developer

  1. Data processing pipelines were experiencing bottlenecks, leading to slow turnaround times. The task aimed to optimize workflows and improve efficiency. Algorithmic optimizations were implemented. Additionally, Solution Modeling was utilized to design and execute a comprehensive improvement plan. This resulted in a 27% enhancement in processing speeds, significantly reducing turnaround times, and improving overall team productivity.2. Traditional Python methods slowed data analysis, hindering timely insights generation. The pipeline was optimized for speed using vectorized operations (NumPy), asynchronous loading (multiprocessing), and memory-mapped files (mmap) for efficient data access. This resulted in a 46% processing time reduction.3. Computationally expensive machine learning models hampered timely analysis in astrophysics research. Optimized algorithms for faster training and prediction by employing efficient models (linear, decision trees), hyperparameter tuning, data pre-processing, and NumPy vectorization. This resulted in a 34% efficiency boost.4. Standard Python methods bottlenecked processing of large FITS files from diverse satellite sensors (X-ray, thermal). Implemented efficient data handling (astropy.io.fits, scikit-image) with selective loading, chunking, & NumPy vectorization for faster processing. Explored CUDA & HPC parallelization for potential hardware acceleration. cProfile-guided code optimization achieved a 65% efficiency gain, enabling faster 2D image generation of diverse satellite imagery. Show less

• 

  University of Michigan-Dearborn

  Sept 2024 - now

  Research Assistant

  • Pioneering Deepfake Audio Authentication Research: Spearheaded advanced research in deepfake audio authentication, successfully enhancing model performance by 15% through innovative applications of spectral analysis and adaptive thresholding techniques. This approach utilized intricate analysis of audio spectra to differentiate authentic audio from deepfake manipulations, significantly improving the robustness and reliability of audio verification.• Designing Neural Networks for Enhanced Spectral Analysis: Engineered and deployed custom neural network architectures tailored for deepfake detection in audio, which increased detection accuracy by 20%. By leveraging multi-network outputs and optimizing network layers specifically for spectral data, this solution enabled more precise detection of subtle audio anomalies, contributing to breakthroughs in audio deepfake identification. Show less