CS @ Penn State. I work on reliability and safety of AI/ML systems. I keep an OSS contribution log.
- Warren — Turn your Wikipedia rabbit hole into a shareable map, with LLM-generated connective tissue explaining how each article follows from the last.
- match-loss-to-cost — Decision-aware traffic forecasting with asymmetric losses and conformal capacity bands. Submitted to CNSM 2026. Evaluated on Abilene, GÉANT, and CESNET-TimeSeries24 across LSTM, DLinear, iTransformer, PatchTST, and DCRNN.
- SecureCodeRL — Scheduled Partial Credit RL for Reliable Code Generation with Small Language Models. LCTES '26 WIP. A binary→partial-credit curriculum lifts syntax validity to 63% on APPS+ stdin prompts, where binary-reward PPO regresses below the SFT baseline.
- Performance-Bugs-LLM — Detecting and explaining Java performance bugs with a fine-tuned LLM. 490-bug dataset over 17 Defects4J projects. IEEE AITest 2025.
- Shift — AI-personalized sustainability actions with radical transparency about AI's environmental cost.
- TruthCast — Multi-agent fact-checking pipeline with on-chain provenance on Solana.