Quinn Arnold

Machine Learning & Applied Mathematics

Researcher · Bryant University · Advanced Analytics Intern

I work on reinforcement learning, deep learning, responsible ML, optimization, and algorithm engineering. Current work spans recurrent PPO for partially observed combinatorial auctions, prompt-injection defense for LLM systems, ML model governance and LLM workflows, and route-optimization research.

  • FastCombo — Deep RL for Combinatorial Auctions

    Recurrent PPO · partially observed auctions · current Bryant collaboration

    A recurrent policy maps public CATS auction histories directly to item prices. It reached 92.4% deterministic exact stage-1 clearing on Paths, above the published 88% Bayesian result under a different protocol.

  • Ask Tupper — Hardened Campus Chatbot

    LLM safety · Qwen3-32B · 2026

    QLoRA fine-tune with a six-layer prompt-injection defense pipeline, reducing attack success from 20% to 0% across 15 tested vectors. RoBERTa classifier achieves 95.3% injection recall.

  • RAPTOR: Route-Audited Patrol Territory Optimization

    Submitted to ALENEX 2027 · graph partitioning · integer staffing

    Routing-aware graph partitioning with integer staffing and audit gates, producing route-audited district geometries and officer allocation recommendations across Providence, Boston, and Seattle street-network instances.

All research & projects →
  • MAPFRE Insurance

  • GuidePost Technologies LLC

  • Rhode Island Novelty

  • SafeWalk

  • Reinforcement learning
  • Deep learning
  • LLM safety
  • Responsible ML evaluation
  • Decision-focused learning
  • Optimization
  • Graph algorithms
  • Game theory
  • Autodiff & backpropagation
  • Applied ML systems