Evaluation, evidence, and real-world data.
My research focuses on problems where system behavior changes over time, including multi-turn language-model safety, financial signals, and market activity.
Dynamic multi-turn youth-safeguarding benchmark
This research examines how conversational risks for young people can evolve across multiple turns. The benchmark is designed to test contextual guardrails beyond single-prompt safety checks.
- Design a multi-turn scenario
- Run the conversation
- Review contextual risks
- Compare guardrail behavior
The evaluation traces how risk changes across a conversation.
Multi-asset meta-labeling pipeline
During summer 2026, I researched a multi-asset portfolio meta-labeling pipeline in collaboration with State Street Global Advisors through Brandeis School of Business and Economics. The work explored portfolio data, time series, and risk modeling.
- Multi-asset data
- Time-series features
- Meta-labeling research
- Risk review
The study connects time-series features to portfolio risk review.
Prediction-market analytics
Through PolyGains, I work with order-book activity, transaction flows, wallet behavior, and trading patterns in decentralized prediction markets.
Test, observe, refine
Strong research turns questions into measurable experiments, records what happens, and improves the next test.