Research · three papers
Models that admit the world changes.
Two published, one under review. Each one started as a question I could not answer from a textbook, and each is linked to the paper, the code or the recorded talk.
Move to shift the regime
Published
Grade 11
Regime-Aware Portfolio Optimization Using an AI–Markov Hybrid Model with Multi-Divergence Learning
International Journal for Multidisciplinary Research (IJFMR), Vol. 8, Issue 2, 2026 · Srijit Mukherjee, AI PhD student, Penn State; supported by Cheenta Academy
A portfolio model that knows which market it is in before it decides what to hold.
1.87 Sharpe ratio, against an equal-weight baseline
260% improvement in Sharpe over equal weight
Published
Grade 12
Predicting Rare Earth Supply Shocks Using a Machine Learning Classifier Built from Open Geopolitical Data
Enterprise Development and Microfinance, Vol. 36, No. 3 (ABDC-indexed) · Dr Loonkar, visiting faculty, NMIMS
A classifier built only from free data that flagged both 2025 Chinese rare-earth restrictions about four weeks before markets moved.
0.87 recall on unseen 2025 data
0.732 ROC-AUC
Under review
Grade 10
Algorithmic Credit Scoring and Disability-Correlated Exclusion: Evidence from a Large-Scale Credit Dataset with Implications for Microfinance
Submitted to Andover Economics Review
Does a credit-scoring model quietly turn away profiles that look like disability? On 1.29 million loans, yes, by 2.8 points.
1.29M loan records
2.8 pts approval gap for the proxy group
02 How I work
Real data, honest error bars.
Three habits across all three papers.
Free, public data
GDELT, Lending Club, ETF prices, state media. If the data costs money, the result cannot be checked by the people it is about.
No leakage
Chronological train and test splits, chi-square checks on model assumptions, reweighting to rule out sampling effects.
Report the failure
The rare-earth classifier raises a false alarm every two trading days. That number is in the abstract, not a footnote.