Hi, I'm
Manish Ram
Chief Scientist
at Parafield
I specialize in Machine Learning in Health, Reinforcement learning, Spatial Audio Generation, and
Building World Models
My Projects
Projects that I Love Building
TL;DR
Python
PyTorch
AI/ML
Undergraduate Researcher
DAIS (UW)
2024 - Current
Undergraduate Researcher
DAIS (UW)
2024 - Current
Student Reseacher
MedArc
2025 - Current
Student Reseacher
MedArc
2025 - Current
IT - Lab Lead
Bellevue College
2023 - Current
IT - Lab Lead
Bellevue College
2023 - Current
Work Experience
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Skills & Tools
ML and SWE Tools
PyTorch
PyTorch
PyTorch
ML Flow
ML Flow
Branding
Branding
SHAP
SHAP
Apache Spark
Apache Spark
LangChain
LangChain
SQL
C/C++
Computer Vision
DeepLearning
Contrastive Learningg
CUDA
CI/CD Pipelines
Hi, I'm
Manish Ram
Chief Scientist
at Parafield
I specialize in Machine Learning in Health, Reinforcement learning, Spatial Audio Generation, and
Building World Models
My Projects
Projects that I Love Building
TL;DR
Python
PyTorch
AI/ML
Undergraduate Researcher
DAIS (UW)
2024 - Current
Undergraduate Researcher
DAIS (UW)
2024 - Current
Student Reseacher
MedArc
2025 - Current
Student Reseacher
MedArc
2025 - Current
IT - Lab Lead
Bellevue College
2023 - Current
IT - Lab Lead
Bellevue College
2023 - Current
Work Experience
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Skills & Tools
ML and SWE Tools
PyTorch
PyTorch
PyTorch
ML Flow
ML Flow
Branding
Branding
SHAP
SHAP
Apache Spark
Apache Spark
LangChain
LangChain
SQL
C/C++
Computer Vision
DeepLearning
Contrastive Learningg
CUDA
CI/CD Pipelines
Hi, I'm
Manish Ram
Chief Scientist
at Parafield
I specialize in Machine Learning in Health, Reinforcement learning, Spatial Audio Generation, and
Building World Models
My Projects
Projects that I Love Building
TL;DR
Python
PyTorch
AI/ML
Undergraduate Researcher
DAIS (UW)
2024 - Current
Undergraduate Researcher
DAIS (UW)
2024 - Current
Student Reseacher
MedArc
2025 - Current
Student Reseacher
MedArc
2025 - Current
IT - Lab Lead
Bellevue College
2023 - Current
IT - Lab Lead
Bellevue College
2023 - Current
Work Experience
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
DAIS - Undergraduate Researcher
• Implemented diffusion transformer (DiT) architectures in PyTorch and TensorFlow to denoise and reconstruct 3D cryo-EM density maps, improving voxel-level fidelity by 15% over baseline DDPM approaches. • Integrated self-attention mechanisms within the reverse diffusion pipeline to better capture long-range dependencies in volumetric data. • Scaled DiT model training across multi-GPU environments using PyTorch Distributed Data Parallel (DDP), reducing training time by 40% while maintaining model convergence. • Designed and optimized custom beta schedules (linear, cosine, and quadratic) for diffusion training, leading to faster convergence and more stable reconstructions
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Med Arc - Student Reseacher
• Contributing to fMRI foundation models using masked autoencoding to learn Image Embeddings from brain imaging data using ViT (Vision Transformers) with data preprocessing, normalization, and augmentation • Implemented network masking and visualization techniques for the Yeo et al. (2011) 7 resting-state networks, enhancing interpretability of brain activity patterns • Developed Inverse Block Masking strategies for brain modeling to improve spatial context capture. • Built lightweight debugging and pretraining scripts optimized for CPU, enabling rapid performance checks and visualization of training dynamics
Skills & Tools
ML and SWE Tools
PyTorch
PyTorch
PyTorch
ML Flow
ML Flow
Branding
Branding
SHAP
SHAP
Apache Spark
Apache Spark
LangChain
LangChain
SQL
C/C++
Computer Vision
DeepLearning
Contrastive Learningg
CUDA
CI/CD Pipelines