Amazon, Seattle, WA, USA Applied Scientist II Intern, Buyer's Risk Prevention (BRP) · May 2026 – Aug. 2026
Working with the Amazon Buyer's Risk Prevention team on LLM agents and buyer fraud.
Paper accepted at the REALM workshop @ EMNLP 2026 — a session-level study of whether LLM shopping agents are behaviorally distinguishable from human shoppers.
Intuit, Mountain View, CA, USA AI Science Intern, QuickBooks AI Science Team · May 2025 – Aug. 2025
Optimized the latency of an agentic workflow (LangGraph) through high-fidelity automated prompt optimization.
Built an automated evaluation suite to assess LLM response fidelity for insights on large-scale consumer data.
Rochester Institute of Technology, Rochester, NY Graduate Research Assistant · Aug. 2023 – Present
PhD student advised by Prof. Ashiqur R. KhudaBukhsh; research focus on responsible and equitable AI systems.
Published in A* venues (ACL, EMNLP, AAAI, IJCAI); research covered by ABC News, CNN, and WIRED.
Mentored three undergraduate researchers through the NSF REU 2026 program — Abhidipto Chakraborty (UT Austin) , Stephanie Birkelbach (Texas A&M) , and Jollen Dai (UCLA) — guiding them from problem formulation to experiments and writing; the work led to two research projects currently under review at top venues.
A math research agent that counts a result as proved only when the Lean 4 compiler accepts the proof. A local LLM tests small cases in Python and formalizes each claim in Lean, Goedel-Prover-V2 searches for proofs, and a reviewer translates every verified statement back into English to catch proofs of an easier claim. It hunts for counterexamples before proving, grounds lemma names with retrieval over 265k Mathlib declarations, steers effort with MAP-Elites, UCB1 bandits, and an island genetic algorithm, and grows a reusable library of verified lemmas. It runs fully local and includes a fine-tuned Goedel-Prover-V2-32B, trained on its own verified proofs over a larger set of ~20k competition problems and research-level lemmas from past open problems; the first round raised pass@1 by 11.9 points on held-out competition problems and 7.5 on miniF2F.
Into formal maths and theorem proving? Check it out on GitHub — feedback and stars welcome! 🚀
Trade-System · Personal Project
A local-first quantitative equity research & paper-trading framework in Python — blending my love of maths, optimization, and machine learning. It ships walk-forward & combinatorial-purged cross-validation with mandatory leakage tests, long-horizon forecasters with conformalized price bands, a nonlinear-dynamics layer (Hurst/fractal, entropy, Lyapunov, RMT, LPPLS bubble detection), and Kelly + Hierarchical Risk Parity position sizing.
Amazon, Seattle, WA, USA
Intuit, Mountain View, CA, USA
Rochester Institute of Technology, Rochester, NY