Washington D.C. · George Washington University
MS in Data Science · he/him
About
I'm Sayan Patra — a data science professional with a strong mathematical foundation and hands-on expertise in machine learning, financial modelling, and business strategy.
My research revolves around Diffusion Models, PINNs, Manifold Dimension Estimation, Stochastic Differential Equations, and Topological Data Analysis.
Currently open to research collaborations — especially on Diffusion Model Enhancements and MDE. Contact me directly.
Master of Science in Data Science
Bachelor of Science (Hons) in Mathematics
Research
Ideas in progress — papers I'm writing, questions I can't stop asking.
Graduate research investigating hybrid architectures combining diffusion generative priors with ResNet50 feature extractors for improved image synthesis quality and training stability. Exploring convergence under stochastic noise schedules and manifold structure in latent space.
Building a Python package for Generalised Partial Autocorrelation — extending classical PACF methods to nonlinear and non-stationary time series settings for more robust signal analysis.
Applying diffusion-based state representations to reinforcement learning environments, aiming to improve policy stability and sample efficiency in high-dimensional state spaces.
Applying persistent homology and TDA tools to high-frequency trading data, identifying structural patterns invisible to classical statistical methods and building persistence diagrams as features for downstream classifiers.
Technical Skills
Work
Public repos = completed work. Private repos = in-progress research. Toggle visibility below.
Click "Edit Visibility" to show hide/show buttons on each project card.
Graduate Research · Diffusion Models
Diffusion Models + ResNet50 implementation. Hybrid generative architecture investigating improved image synthesis quality, training convergence under stochastic noise schedules, and manifold structure in latent space.
Public · CompletedNLP · Political Rhetoric
Final NLP project analyzing political rhetoric patterns across US presidential speeches. Sentiment evolution, topic modelling, and linguistic fingerprinting.
Public · CompletedDeep Learning · MIT 6.S191
Lab materials and projects from MIT 6.S191: Introduction to Deep Learning — completed in-person. Covers CNNs, RNNs, GANs, reinforcement learning, and beyond.
Public · CompletedComputer Vision · Satellite · Abstract
Geospatial cloud detection using diffusion-based image generation combined with satellite imagery. Updated 2 days ago — active research.
Private · In ProgressVisualization · Tableau · Abstract
Research on Visualisation — used platform: Tableau. Graduate level. 51 commits in February 2026 alone. 1 open issue.
Private · In ProgressRobotics · Research · Abstract
Research-level neural architecture for robotics control systems. Bridging deep learning representations with physical system constraints. 17 commits this month.
Private · In ProgressFinance · RL · Abstract
An autonomous trading agent leveraging reinforcement learning and quantitative signals. Connects financial domain expertise with modern RL methods.
Private · In ProgressDiffusion · RL · Abstract
Diffusion-based state representations for sequential decision-making. GPL v3 licensed research project exploring policy stability and sample efficiency.
Private · In ProgressStatistics · Python Package · Abstract
A package for Generalised Partial Autocorrelation extending classical PACF methods to nonlinear and non-stationary time series settings.
Private · In ProgressNLP · Healthcare · Abstract
Chatbot based on Trauma Informed Care principles. Collaborative project with @ichaudh bridging NLP and healthcare communication design.
Private · In ProgressSide Quests
Not everything needs to ship. Sometimes the best learning starts with play.
A model that generates Spotify playlists from facial expression via webcam and transfer learning. Surprisingly accurate at detecting "I should be studying" energy.
Explore →A Telegram bot that summarizes any ArXiv paper into a structured 5-point thread. Because reading 40 pages daily is not a sustainable research strategy.
Try it →Predictive modelling on real estate data — property price estimation, neighbourhood clustering, and investment signal detection using ensemble methods.
GitHub →Projects completed in MIT 6.S191 in-class sessions covering sequence models, music generation, facial recognition, and reinforcement learning game agents.
GitHub →Experience
GitHub
Open to collaboration — especially on Diffusion Model Enhancements and MDE.
github.com/Sayanpatraa · Washington DC · he/him
Actively seeking collaborators. Contact me directly.
Building tools that help the ML community. PRs welcome.
Open to co-writing papers on ML theory, NLP, TDA, or causal inference.
GWU Graduate IA. Happy to help early-career data scientists.
Beyond Work
References
Academic and professional references available on request. Update the names below with your actual references.
Faculty · The George Washington University — Data Science
"Sayan brings rare mathematical depth to applied ML problems. His ability to connect theoretical frameworks like TDA and SDEs to practical implementations makes him a standout research collaborator."
The George Washington University · Washington D.C.
Instructor · Machine Learning I — GWU
"As a Graduate IA, Sayan consistently helped students bridge the gap between theory and implementation. His clarity on gradient descent and model evaluation was the best I've seen from a student assistant."
The George Washington University · Machine Learning I
Senior Manager · Amazon EMEA
"Sayan's analytical approach to eCommerce data stood out immediately. He consistently translated complex transaction patterns into actionable business strategy that the team actually implemented."
Amazon · EMEA Market Division
Portfolio Manager · ICICI Securities Private Limited
"A meticulous, data-driven mind in equity analysis. Sayan's ability to synthesize market signals with portfolio risk models — well beyond his years of experience — impressed our entire advisory team."
ICICI Securities Private Limited · India
Contact
"Turning data into decisions, and decisions into meaningful change."
EN · 中文 · ES · FR · DE