//Resume

Owned trace generation for post-training: built the pipeline producing human-verified agentic trajectories used as SFT data for an open-weights LLM, and contributed to the broader post-training effort (SFT, distillation, LoRA) on Google Cloud. Built and maintained the LLM evaluation framework, a human-verified agentic benchmark with full LangSmith tracing, used to compare frontier models; trace-level failure analysis drove targeted fixes that improved correctness across models. Re-architected the production multi-agent harness (LangGraph, LangSmith), reducing token usage and end-to-end latency.

Studying congressional speech effectiveness via ML with Prof. Bryce J. Dietrich. Built a LexisNexis pipeline extracting structured coverage metadata, and implemented a dual-label (human + LLM) tone-classification pipeline.

Website Chair and Organizer for Purdue's largest CS/DS mentorship community. LaunchPad is a one-on-one, semester-long mentorship program that pairs incoming students with upperclassmen, guiding them in building technical projects and fostering community through events and skill-building workshops.

Selected as 1 of 50 technicians from ~15,000 applicants for a 55,000-attendee conference. Tracked and maintained 1,000+ laptops, set up presentation equipment, assisted broadcasters/guests with technical issues, and coordinated daily patches and team memos.

Developed an AI-driven credit-scoring solution integrating bank data with nontraditional signals — mobile-provider behavior, ride-hailing driver earnings, and other alternative data. Built an AI chatbot to explain Payriff service APIs, streamlining developer onboarding.