Shreyarth Thakor

About

B.Tech Information Technology student building data science and quantitative research systems across financial NLP, portfolio optimization, experimentation, and probabilistic modeling. Experienced with end-to-end data pipelines, walk-forward validation, statistical inference, and translating model output into decision-oriented applications.

Education

• 2028 SI
  B.Tech

  Sarvajanik Institute of Engineering and Technology

  Information Technology · Jul 2024 – Jun 2028

  Information Technology · Jul 2024 – Jun 2028

Skills

Tools / apps / platforms

Projects

• Monte Carlo Stock Price Simulation & Backtesting ↗
  Python, NumPy, pandas, matplotlib, Streamlit, Alpha Vantage

  Simulated 1K-10K price paths using Geometric Brownian Motion with Student-t shocks for fat-tailed returns and validated forecasts through walk-forward evaluation across 30+ historical periods. Built an interactive Streamlit application with Alpha Vantage integration for Indian-market tickers, surfacing price distributions, confidence intervals, probability-of-profit metrics, and backtest diagnostics.

• A/B Testing Intelligence Platform ↗
  Python, SciPy, pandas, NumPy, Streamlit, Plotly

  Built an interactive experimentation platform covering t-tests, chi-square and Mann-Whitney U, verified against SciPy, with safeguards for Simpson's Paradox, multiple testing, and sequential peeking using Bonferroni and Benjamini-Hochberg corrections. Translated statistical outputs into incremental revenue, annualized lift, ROI, and ship/don't-ship recommendations through an executive reporting layer.

• Multi-Strategy Behavioral Portfolio Optimization & Backtesting ↗
  Python, NumPy, pandas, SciPy, scikit-learn, yfinance

  Built a portfolio research engine across 30 consumer-behavior equities benchmarked against SPY; walk-forward tested 6 allocation methods using expanding-window Ledoit-Wolf covariance estimation and annual rebalancing. 3 of 6 strategies beat SPY out-of-sample (2021-2025), led by Max Sharpe at ~16% CAGR vs. SPY's ~15%; stress-tested concentration and reported when apparent outperformance disappeared after excluding a dominant position.

• Corporate Misconduct Early Warning System ↗
  Python, pandas, scikit-learn, XGBoost, NLP, FinNLP

  Built a memory-efficient financial-NLP research pipeline over 68,874 SEC 10-K firm-year filings, joining MD&A text with AAER enforcement labels and engineering Loughran-McDonald and text-surface risk features for a fixed 5% analyst-review queue. Detected temporal leakage affecting 30,582 of 41,748 development rows; rebuilt the pipeline with strict filing-date walk-forward validation, zero filing-year mismatches or fold overlap, plus regression, artifact-integrity, threshold-selection, and sealed-test safeguards.