I’m an applied mathematics and statistics researcher working across reinforcement learning, deep learning, LLM safety and evaluation, responsible ML, graph optimization, and production AI systems.
I’m completing a B.S. in Applied Mathematics & Statistics at Bryant University, expected December 2026.
My current research includes recurrent PPO for partially observed combinatorial auctions, route optimization under operational constraints, and rigorous evaluation for LLM and ML systems. I’m especially interested in connecting experimental results to decisions and systems that can be audited in practice.
Open to research scientist and applied ML opportunities, including AI-lab roles focused on reinforcement learning, deep learning, and reliable AI.
- FastCombo — Deep RL for Combinatorial Auctions: In a current collaboration with Bryant faculty and a graduate student, a recurrent PPO policy maps public histories from partially observed CATS auctions directly to prices. It reached 92.4% deterministic exact stage-1 clearing on Paths, compared with a published 88% Bayesian result under a different protocol.
- Ask Tupper — LLM Safety & Evaluation: Fine-tuned Qwen3-32B with QLoRA and built a six-layer prompt-injection defense pipeline that reduced attack success from 20% to 0% across 15 tested vectors; a custom RoBERTa classifier achieved 95.3% injection recall.
- GuidePost — Routing Research & Decision Support: Across 264 paired Boston conditions, KDE-penalized routes increased retrospective holdout-incident encounters by 26.3 and route coverage by 16.7 percentage points versus shortest-path routes with identical peaks and ordering. The research supports a patrol decision-support platform with authenticated, audit-logged workflows.
- SafeWalk — Open-Source Route Evaluation: Built an MIT-licensed iOS system for comparing shortest and risk-weighted routes. Multicity evaluations with strict temporal holdout compared KDE, XGBoost, and graph neural networks; KDE achieved at least 50% greater corridor-based holdout-incident avoidance in every city.
- RAPTOR — Graph Optimization: Developed route-audited graph partitioning with integer staffing constraints and evaluated candidate district plans on Providence, Boston, and Seattle street networks for a submitted ALENEX 2027 paper.
- MAPFRE — Responsible ML & AI Workflows: Contributing to model governance through bias assessment, mitigation analysis, reliability review, and production-readiness evaluation, while developing LLM-assisted image-analysis workflows for underwriting review.
- ML & research: PyTorch, recurrent PPO, QLoRA, RoBERTa, RAG, XGBoost, model evaluation, bias assessment
- Algorithms & systems: optimization, graph algorithms, KDE, experimental design, Python, SQL, Swift, TypeScript, AWS SageMaker, Docker