I like problems where the mathematics is elegant and the real world refuses to cooperate.
Undergraduate in Mathematics & Computer Science at Springfield College. Interested in optimization, probabilistic modeling, causal inference, and whether our software deserves our trust.
I grew up in Nepal, and now study mathematics and computer science in the Honors Program at Springfield College.
I'm drawn to the assumptions hiding inside proofs, models, and numerical results. Sometimes that means studying tail risk in market mechanisms. Other times it means debugging an optimization library or asking whether a causal model is learning what we think it is.
Outside research I work with campus technology, support students, and collect more interesting questions than I have time to answer.
Problems I have spent a considerable number of evenings thinking about.
How can a trading agent look calibrated on average yet carry catastrophic tail risk? I studied this question for a logarithmic market scoring rule (LMSR) market maker, deriving analytical results and testing position limits and insolvency in reproducible simulations.
The manuscript and simulation code are being prepared for public release.
Examining when balancing penalties can shift treatment-effect estimates toward confounded comparisons, using controlled simulations and ICU time-series data. After finding a timestep-indexing error in an earlier pipeline, I withdrew the affected workshop paper and rebuilt the analysis with frozen splits and stronger validation checks.
Additional controls and external-cohort validation are in progress; revised findings have not yet been published.
At a CMU × NVIDIA hackathon, I helped analyze metadata compatibility across international biomedical sources. Only 120 of 11,511 variables appeared consistently across at least eight of 14 sources. It is a striking gap between the promise of pooled data and what can actually be compared.
Built a stochastic model for a lunar supply chain with random failures and degradation. Compared adaptive and static policies using Monte Carlo simulations in the Mathematical Contest in Modeling.
I contribute to the open-source Python library for convex optimization, especially its correctness and post-solve verification. That includes complex positive-semidefinite constraints, constraint residuals, shape inference, and regression tests.
10merged pull requests
to the core library.
BLAST-based experiments on chromosome-level cluster assignments, bidirectional best hits, and representative selection.
Compared schemes for orbital dynamics to study numerical error growth and stability.
Resident Assistant
Springfield College
Math, Physics & CS Student Assistant
Springfield College
Information Technology Services
Springfield College
Invited Reviewer · ICML Workshop
Foundation Models for Structured Data
Participant
Jane Street Estimathon; Harvard Undergraduate Trading Competition
Research ideas, mathematical questions, open-source issues, or just a good conversation.
adityakarna820@gmail.com