Vishnu Datta Jayanti

// 01 . LOG

SYSTEM OPTIMAL

education

B.S. Computer Science, Purdue University, John Martinson Honors College. Expected graduation May 2030.

focus

ML research, applied ML, and quant backtesting. Focused on building predictive models and systematic strategies grounded in real-world data.

approach

I test things on real data before I trust them. Six projects so far, each one backtested, benchmarked, or peer reviewed in some form.

now

Working on volatility-based trading strategies and looking for research or internship opportunities in ML, quant, and SWE

// 02 . RESEARCH

Publications

May 2026

PUBLISHED

Wharton Sports
Analytics Journal

Beyond the Expert: An Algorithmic Approach to Correcting Expert Bias in Fantasy Football Projections

Original research evaluating whether machine learning can mitigate cognitive bias in fantasy football projections. Position-specific XGBoost models were trained on ten years of NFL data with over 1,900 engineered features. By mitigating cognitive bias, I was able to improve ordinal rankings, especially for quarterbacks and tight ends, in comparison to ESPN.

XGBoost 10yr NFL data 1,900+ features bias correction

VIEW PAPER ↗

// 03 . BUILDS

Engineering

MOD-01

QUANT_ENGINE

Apr – Jul 2026 Python · Pandas · NumPy

Volatility-Adjusted Momentum Engine

A point-in-time backtesting framework for U.S. equities. The strategy identifies price breakouts via 50/200-day moving average spreads, normalizes signals using Garman-Klass volatility, and allocates capital inversely to risk. Spanning seven pipeline stages, the daily-rebalanced execution simulator features strict liquidity guardrails and dynamic equity scaling, where positions are sized proportionally to total net worth in order to maximize compounding.

$70K → $285simulated portfolio, 12yr backtest

2012–2024historical coverage

point-in-timeno look-ahead bias

ACCESS REPO ↗

MOD-02

ML_PIPELINE

Aug 2025 - May 2026 Python · XGBoost · scikit-learn · Pandas

NFL Expert Bias Mitigation

The functional implementation of my published research: a tool that generates fantasy football rankings from position-specific XGBoost models and compares them directly against ESPN's expert projections to uncover undervalued and overvalued players. Includes a full evaluation pipeline that scores both projections against final season results to see who was actually right.

4 modelsQB, RB, WR, TE

2013-2023historical dataset

1900+features utilized

ACCESS REPO ↗

MOD-03

SYS_PROG

Feb – Apr 2026 C

VaultCLI

A command-line password manager written in C from the ground up. Master-password authentication gates a vault encrypted with a masster-password based XOR cipher, stored in custom binary structs via raw file I/O. Includes duplicate detection, strength scoring across length and character variety, and randomized 20-character password generation.

custombinary storage format

XOR cipherbuilt from scratch

zeroexternal crypto libraries

ACCESS REPO ↗

MOD-04

RAG_SYS

Oct 2024 – Aug 2025 Python · Flask · XGBoost · LangChain · NBA API

The Basketball Oracle

An NBA analytics platform with live scores, standings, rosters, and boxscores, plus a machine-learning core that predicts player performance for the upcoming season. A toggle switches between fast stat lookups and an AI search mode: a LangChain and Ollama RAG pipeline that answers natural-language basketball questions over a Wikipedia-derived knowledge base.

0.71 R²avg across 5 NBA metrics

52,000+indexed articles (RAG)

KMeans + XGBoostprediction core

ACCESS REPO ↗

MOD-05

NLP_ENGINE

Jun – Aug 2025 Python · NLTK · WordNet

Semantic Frame Identifier

An NLP system that extracts three semantic frames, capital stock, commercial transaction, and business, from financial text. Named entities are corrected with a custom gazetteer, frames are matched against lexical units using Wu-Palmer and Leacock-Chodorow WordNet similarity scoring, and frame elements like buyer, seller, and monetary amount are pulled from surrounding context.

3 framesstock, transaction, business

WordNetsimilarity scoring

customgazetteer + NER correction

ACCESS REPO ↗

MOD-06

FULL_STACK

Oct 2024 - Apr 2025 Flask · Python · JavaScript · XGBoost

Differential Diagnosis

A full-stack web app that predicts likely diseases from user-reported symptoms, built with a small team. I led the project, trained the XGBoost classifier behind the prediction engine, and built the Flask backend along with a health-history tracker that lets logged-in users review past diagnoses over time, all behind secure, hashed authentication.

XGBoostsymptom → disease classifier

3-personteam, led by me

deployedlive on Render

ACCESS REPO ↗ ACCESS WEBSITE ↗

// 04 . TECH STACK

System Loadout

LANGUAGES

Python
C
HTML
CSS

ML & DATA

XGBoost
Scikit-learn
Pandas
NumPy
NLTK

INFRA

Flask
LangChain
Git / GitHub

// 05 . CONTACT

Reach Out.

Open to research collaborations, internships, and interesting projects.

EMAIL vishnudattaj@gmail.com LINKEDIN /in/vjayanti GITHUB /vishnudattaj SCHOLAR Google Scholar ↗