NEWVisual Study Atlas: open-source Claude Code skill that turns books into offline study atlases · watch the 8.5-min walkthrough →
Academic project disclaimer: This website and the projects linked here are academic and personal learning projects only. They are not investment advice, trading advice, or any recommendation to buy, sell, or hold securities or financial instruments. Views expressed are strictly personal and are not to be construed as the views, policy, endorsement, or advice of my organization or employer.
// banking · risk · forecasting · open source

Sabyasachi Chowdhury

MS Applied Data Science · UChicago Bayesian Methods · Financial Risk · ML

I am a banker and data scientist by passion, building AI-driven decision-support systems for financial supervision, macro-financial risk, and probabilistic forecasting. My work combines Bayesian methods, machine learning, causal inference, and stochastic modeling to support decisions under uncertainty. I build in the open: a causal-diagnostics workbench for trading ideas, probabilistic market dashboards, and Visual Study Atlas, an open-source Claude Code skill that turns books into offline study atlases.

★ Featured: Visual Study Atlas ↓ View Projects ✉ Get in touch ⌥ GitHub
Picture Says a Thousand Words: one question, six tabs ▶
🖼️● open source
Picture Says a Thousand Words

A Claude Code skill that turns one question into one offline tabbed page: animation, interactive micro-app, diagram, clickable flow map, a first-principles derivation and a short recap, with typeset maths and an optional local tutor chat. Built from the ladder of formats in a public post by Andrej Karpathy (independent project, not affiliated).

Claude CodeKaTeXInteractiveOfflineMIT
▶ Project page & video Live demo page ⌥ GitHub
AI Systems for Risk & Forecasting
🧭 Causal Strategy Harness ↗ ● live

An interactive research workbench that turns a plain-English trading idea into an assumption-aware causal diagnostic. Users choose the asset, period, confounders, estimators, and robustness settings; the app then compares the naive backtest with DAG-adjusted estimates, placebo tests, horizon sensitivity, and heterogeneous effects. The final output is a readable evidence report designed to separate a plausible signal from market-regime correlation—without presenting observational evidence as guaranteed alpha.

Streamlit DoWhy EconML Double ML Causal Forest DAG LLM-assisted yfinance
Research workflow
InputStrategy claim
BaselineNaive backtest
AdjustmentCausal battery
Stress testsPlacebo + horizons
OutputProvisional verdict
📊 Macro Market Intelligence Dashboard ↗ ● live

AI-powered macroeconomic forecasting and market-risk platform centered on a probabilistic Forecast Lab. It combines Monte Carlo simulation, Markov regime detection, historical analogue matching, and Bayesian reasoning to estimate a range of future scenarios rather than one fixed prediction, with supporting macro monitoring, cross-asset risk analytics, and news intelligence.

Streamlit FRED API Kraken Polymarket Macro Regime Yield Curve 27 assets
▶ Open live app ⌥ GitHub
📈 India Market Terminal ↗ ● live

Probabilistic forecasting platform for Indian financial markets built around a Forecast Lab that estimates future scenarios using Monte Carlo simulation, Markov regime models, and historical analogue analysis. It complements this with Nifty 500 market breadth, liquidity, volatility, and systemic-risk monitoring.

Streamlit India Markets Nifty 500 yfinance Market Regime Screener Risk Monitor
▶ Open live app ⌥ GitHub
🧬 Prostate MRI PI-RADS Predictor ↗ ● live

Physics-informed deep learning framework for prostate-cancer risk stratification from MRI. A Variational Autoencoder learns clinically meaningful latent features while jointly reconstructing medical images and predicting PI-RADS scores, improving robustness and interpretability. Built at Uncommon Hacks 2026 and deployed on Hugging Face Spaces.

VAE PyTorch Medical AI Gradio HuggingFace NYU fastMRI Hackathon 🏆
▶ Open live app ▷ Demo video ⌥ GitHub
About

I am a banker and currently pursuing an MS in Applied Data Science at the University of Chicago. My work sits at the intersection of Bayesian statistics, machine learning, causal inference, and macro-financial risk.

I approach problems through a probabilistic and Bayesian lens, building models that explain uncertainty and support supervisory, policy, and risk-management decisions rather than simply producing point predictions.

Most of the projects here are open source on GitHub, from causal-inference tooling and market-risk dashboards to learning tools such as Visual Study Atlas. My broader interests include stress testing, early-warning systems, time-series forecasting, reinforcement learning, and financial NLP.

// stack

  • Python / NumPy
  • Causal Inference
  • ML / EconML
  • Time Series
  • Bayesian Methods
  • Quant Finance
Contact
✉ sabyachow@gmail.com ⌥ github.com/sabya-chow in LinkedIn
Academic project disclaimer: This website and the projects linked here are academic and personal learning projects only. They are not investment advice, trading advice, or any recommendation to buy, sell, or hold securities or financial instruments. Views expressed are strictly personal and are not to be construed as the views, policy, endorsement, or advice of my organization or employer.